use train_loss for sp test

* synthetic datasets for benchmarking and testing

* fix synthetic dataset parse from config and add tests

* use type=_synthetic

_quarto.yml

@@ -128,11 +128,9 @@ quartodoc:
         - monkeypatch.mistral_attn_hijack_flash
         - monkeypatch.multipack
         - monkeypatch.relora
-        - monkeypatch.llama_expand_mask
         - monkeypatch.lora_kernels
         - monkeypatch.utils
         - monkeypatch.btlm_attn_hijack_flash
-        - monkeypatch.llama_patch_multipack
         - monkeypatch.stablelm_attn_hijack_flash
         - monkeypatch.trainer_fsdp_optim
         - monkeypatch.transformers_fa_utils

benchmarks/bench_scattermoe_lora.py

@@ -0,0 +1,284 @@
+"""Benchmark for ScatterMoE LoRA Triton kernels.
+
+Measures forward, backward dX, and backward dA/dB kernels at common MoE
+model shapes. Reports per-kernel timings, LoRA overhead vs base scatter2scatter,
+and full fwd+bwd autograd throughput.
+
+Usage:
+  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py
+  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py --ranks 16 64
+  CUDA_VISIBLE_DEVICES=0 python benchmarks/bench_scattermoe_lora.py --models Qwen/Qwen3.5-35B-A3B
+"""
+
+import argparse
+import gc
+import time
+from functools import partial
+
+import torch
+
+from axolotl.integrations.kernels.libs.scattermoe_lora.kernels import (
+    lora_ops,
+    ops as base_ops,
+)
+from axolotl.integrations.kernels.libs.scattermoe_lora.parallel_experts import (
+    flatten_sort_count,
+)
+from axolotl.integrations.kernels.libs.scattermoe_lora.parallel_linear_lora import (
+    ScatterMoELoRA,
+)
+
+DEVICE = "cuda"
+DTYPE = torch.bfloat16
+WARMUP = 5
+ITERS = 20
+
+# ─── Model configs ──────────────────────────────────────────────────────────
+
+BUILTIN_CONFIGS = {
+    "Qwen3.5-35B-A3B": (256, 2048, 512, 8),  # E, H, I, k
+    "Qwen3-30B-A3B": (128, 2048, 768, 8),
+    "OLMoE-1B-7B": (64, 2048, 1024, 8),
+    "Mixtral-8x7B": (8, 4096, 14336, 2),
+}
+
+
+def _resolve_config(spec):
+    """Resolve a model spec to (E, H, I, k). Accepts builtin names or HF IDs."""
+    key = spec.lower().replace("/", "-")
+    for name, cfg in BUILTIN_CONFIGS.items():
+        if key in name.lower() or name.lower() in key:
+            return name, cfg
+
+    from transformers import AutoConfig
+
+    hf_cfg = AutoConfig.from_pretrained(spec, trust_remote_code=True)
+    if callable(getattr(hf_cfg, "get_text_config", None)):
+        tc = hf_cfg.get_text_config()
+        if hasattr(tc, "model_type") and tc.model_type != hf_cfg.model_type:
+            hf_cfg = tc
+    hidden = hf_cfg.hidden_size
+    inter = getattr(hf_cfg, "moe_intermediate_size", None) or hf_cfg.intermediate_size
+    experts = (
+        getattr(hf_cfg, "num_experts", None)
+        or getattr(hf_cfg, "num_local_experts", None)
+        or getattr(hf_cfg, "n_routed_experts", None)
+    )
+    top_k = (
+        getattr(hf_cfg, "num_experts_per_tok", None)
+        or getattr(hf_cfg, "num_experts_per_token", None)
+        or 2
+    )
+    name = spec.split("/")[-1]
+    return name, (experts, hidden, inter, top_k)
+
+
+# ─── Benchmark helpers ──────────────────────────────────────────────────────
+
+
+def _clean():
+    gc.collect()
+    torch.cuda.empty_cache()
+    torch.cuda.synchronize()
+
+
+def _bench(fn, warmup=WARMUP, iters=ITERS):
+    for _ in range(warmup):
+        fn()
+    torch.cuda.synchronize()
+    times = []
+    for _ in range(iters):
+        torch.cuda.synchronize()
+        t0 = time.perf_counter()
+        fn()
+        torch.cuda.synchronize()
+        times.append((time.perf_counter() - t0) * 1000)
+    times.sort()
+    return times[len(times) // 2]
+
+
+def _setup(num_experts, K, N, T, top_k, R):
+    torch.manual_seed(42)
+    x = torch.randn(T, K, device=DEVICE, dtype=DTYPE)
+    W = torch.randn(num_experts, K, N, device=DEVICE, dtype=DTYPE) * 0.02
+    lora_A = torch.randn(R * num_experts, K, device=DEVICE, dtype=DTYPE) * 0.01
+    lora_B = torch.randn(N, R * num_experts, device=DEVICE, dtype=DTYPE) * 0.01
+    logits = torch.randn(T, num_experts, device=DEVICE)
+    _, top_idx = torch.topk(torch.softmax(logits, dim=-1), top_k, dim=-1)
+    sei, ssi, eo = flatten_sort_count(top_idx, num_experts)
+    gx = base_ops.group(x, ssi, fan_out=top_k)
+    dy = torch.randn(gx.size(0), N, device=DEVICE, dtype=DTYPE)
+    return x, W, lora_A, lora_B, sei, ssi, eo, gx, dy
+
+
+# ─── Kernel wrappers (avoid B023 loop-variable capture) ──────────────────────
+
+
+def _call_fwd(x, W, sei, ssi, top_k, lA, lB):
+    return lora_ops.scatter2scatter_lora(
+        X=x,
+        W=W,
+        sorted_expert_idxs=sei,
+        sorted_scattered_idxs=ssi,
+        k=top_k,
+        lora_A=lA,
+        lora_B=lB,
+        scaling=2.0,
+    )
+
+
+def _call_base(x, W, sei, ssi, top_k):
+    return base_ops.scatter2scatter(
+        X=x,
+        W=W,
+        sorted_expert_idxs=sei,
+        sorted_scattered_idxs=ssi,
+        k=top_k,
+    )
+
+
+def _call_dx(dy, W, sei, ssi, lA, lB):
+    return lora_ops.scatter2scatter_lora_dX(
+        DY=dy,
+        W=W,
+        sorted_expert_idxs=sei,
+        sorted_scattered_idxs=ssi,
+        k=1,
+        lora_A=lA,
+        lora_B=lB,
+        scaling=2.0,
+        dy_grouped=True,
+        dx_grouped=False,
+    )
+
+
+def _call_bwd(dy, gx, lA, lB, eo, num_experts):
+    return lora_ops.group_bwd_lora(
+        DY=dy,
+        X=gx,
+        lora_A=lA,
+        lora_B=lB,
+        expert_offsets=eo,
+        E=num_experts,
+        scaling=2.0,
+    )
+
+
+# ─── Main ────────────────────────────────────────────────────────────────────
+
+
+def main():
+    parser = argparse.ArgumentParser(description="ScatterMoE LoRA kernel benchmark")
+    parser.add_argument(
+        "--models",
+        "-m",
+        nargs="+",
+        help="Model names or HF IDs (default: all builtins)",
+    )
+    parser.add_argument("--ranks", "-r", nargs="+", type=int, default=[16, 32, 64])
+    parser.add_argument("--seq-len", "-T", type=int, default=2048)
+    args = parser.parse_args()
+
+    T = args.seq_len
+    print(f"GPU: {torch.cuda.get_device_name()}")
+    print(f"T={T}, ranks={args.ranks}\n")
+
+    if args.models:
+        configs = [_resolve_config(m) for m in args.models]
+    else:
+        configs = list(BUILTIN_CONFIGS.items())
+
+    for model_name, (num_experts, hidden, inter, top_k) in configs:
+        print(f"{'=' * 70}")
+        print(f"  {model_name}: E={num_experts}, H={hidden}, I={inter}, k={top_k}")
+        print(f"{'=' * 70}")
+
+        for R in args.ranks:
+            for proj, K, N in [("gate_up", hidden, 2 * inter), ("down", inter, hidden)]:
+                _clean()
+                x, W, lA, lB, sei, ssi, eo, gx, dy = _setup(
+                    num_experts, K, N, T, top_k, R
+                )
+
+                # Forward with LoRA (auto-dispatched: fused or split)
+                dispatch = (
+                    "split"
+                    if (
+                        num_experts <= lora_ops._SPLIT_LORA_FWD_MAX_EXPERTS
+                        and K * N >= lora_ops._SPLIT_LORA_FWD_THRESHOLD
+                    )
+                    else "fused"
+                )
+                t_fwd = _bench(partial(_call_fwd, x, W, sei, ssi, top_k, lA, lB))
+                t_base = _bench(partial(_call_base, x, W, sei, ssi, top_k))
+                t_dx = _bench(partial(_call_dx, dy, W, sei, ssi, lA, lB))
+                t_bwd = _bench(partial(_call_bwd, dy, gx, lA, lB, eo, num_experts))
+
+                total = t_fwd + t_dx + t_bwd
+                overhead = t_fwd / t_base - 1 if t_base > 0 else 0
+
+                print(
+                    f"  R={R:>2} {proj:<8}  "
+                    f"fwd={t_fwd:>6.2f}ms [{dispatch}]  "
+                    f"base={t_base:>6.2f}ms "
+                    f"(+{overhead * 100:.0f}%)  "
+                    f"dx={t_dx:>6.2f}ms  bwd={t_bwd:>6.2f}ms  "
+                    f"total={total:>6.2f}ms"
+                )
+
+                # Full autograd fwd+bwd with memory measurement
+                x_ag = x.clone().requires_grad_(True)
+                lA_ag = lA.clone().requires_grad_(True)
+                lB_ag = lB.clone().requires_grad_(True)
+
+                def _run_autograd(
+                    _x=x_ag,
+                    _W=W,
+                    _k=top_k,
+                    _sei=sei,
+                    _ssi=ssi,
+                    _eo=eo,
+                    _lA=lA_ag,
+                    _lB=lB_ag,
+                ):
+                    out = ScatterMoELoRA.apply(
+                        _x,
+                        _W,
+                        _k,
+                        _sei,
+                        _ssi,
+                        _eo,
+                        _lA,
+                        _lB,
+                        2.0,
+                        None,
+                        None,
+                        False,
+                        False,
+                        True,
+                        False,
+                    )
+                    out.sum().backward()
+                    _x.grad = None
+                    _lA.grad = None
+                    _lB.grad = None
+
+                t_full = _bench(_run_autograd)
+
+                _clean()
+                torch.cuda.reset_peak_memory_stats()
+                mem_before = torch.cuda.memory_allocated()
+                _run_autograd()
+                torch.cuda.synchronize()
+                mem_peak = torch.cuda.max_memory_allocated() - mem_before
+
+                print(
+                    f"         full_fwd_bwd={t_full:>6.2f}ms  "
+                    f"peak_delta={mem_peak / 1e6:>6.1f}MB"
+                )
+
+        print()
+
+
+if __name__ == "__main__":
+    main()

cicd/Dockerfile-uv.jinja

@@ -11,7 +11,7 @@ ENV NIGHTLY_BUILD="{{ NIGHTLY_BUILD }}"
 ENV HF_HOME="{{ HF_HOME }}"
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
 
 WORKDIR /workspace
 

cicd/Dockerfile.jinja

@@ -12,7 +12,7 @@ ENV HF_HOME="{{ HF_HOME }}"
 ENV AXOLOTL_DATASET_NUM_PROC="8"
 
 RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
+    apt-get install -y --allow-change-held-packages vim curl nano zstd libnccl2 libnccl-dev ibverbs-providers ibverbs-utils infiniband-diags librdmacm-dev librdmacm1 rdmacm-utils slurm-wlm
 
 WORKDIR /workspace
 

cicd/cicd.sh

@@ -3,11 +3,13 @@ set -e
 
 python -c "import torch; assert '$PYTORCH_VERSION' in torch.__version__"
 
-# curl -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C "${HF_HOME}/hub/"  --use-compress-program unzstd --strip-components=1
-hf download "NousResearch/Meta-Llama-3-8B"
-hf download "NousResearch/Meta-Llama-3-8B-Instruct"
-hf download "microsoft/Phi-4-reasoning"
-hf download "microsoft/Phi-3.5-mini-instruct"
+set -o pipefail
+curl --silent --show-error --fail --retry 3 --retry-delay 5 -L https://axolotl-ci.b-cdn.net/hf-cache.tar.zst | tar -xpf - -C "${HF_HOME}/hub/" --use-compress-program unzstd --strip-components=1
+# hf download "NousResearch/Meta-Llama-3-8B"
+# hf download "NousResearch/Meta-Llama-3-8B-Instruct"
+# hf download "microsoft/Phi-4-reasoning"
+# hf download "microsoft/Phi-3.5-mini-instruct"
+# hf download "microsoft/Phi-3-medium-128k-instruct"
 
 # Run unit tests with initial coverage report
 pytest -v --durations=10 -n8 \

cicd/single_gpu.py

@@ -68,10 +68,6 @@ def run_cmd(cmd: str, run_folder: str):
     sp_env["AXOLOTL_DATASET_NUM_PROC"] = "8"
 
     # Propagate errors from subprocess.
-    try:
-        exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
-        if exit_code:
-            print(f"Command '{cmd}' failed with exit code {exit_code}")
-            return exit_code
-    except Exception as e:  # pylint: disable=broad-except
-        print(f"Command '{cmd}' failed with exception {e}")
+    exit_code = subprocess.call(cmd.split(), cwd=run_folder, env=sp_env)  # nosec
+    if exit_code:
+        raise RuntimeError(f"Command '{cmd}' failed with exit code {exit_code}")

codecov.yml

@@ -37,6 +37,7 @@ coverage:
         only_pulls: false
         flags: null
         paths: null
+        informational: true
 
 parsers:
   gcov:

docs/gradient_checkpointing.qmd

@@ -1,5 +1,5 @@
 ---
-title: Gradient Checkpointing and Activation Offloading
+title: Gradient Checkpointing, Activation Offloading, and Layer Offloading
 ---
 
 Gradient checkpointing and activation offloading are techniques used to optimize the performance of deep learning
@@ -27,3 +27,33 @@ The `activation_offloading: legacy` naively offloads activations to CPU and with
 
 For resource constrained environments with limited CPU memory, `activation_offloading: disk` offloads
 activations to disk instead of CPU RAM so that much larger context lengths can be trained with minimal memory.
+
+### Enabling Layer Offloading
+
+```yaml
+layer_offloading: true
+```
+
+Layer offloading reduces GPU memory usage by moving frozen (non-trainable) decoder layer parameters to CPU
+and streaming them back to GPU one layer at a time during the forward and backward passes. This is
+particularly useful for LoRA/QLoRA training where most of the model's parameters are frozen — only the
+trainable adapter weights stay on GPU permanently.
+
+During training, forward and backward hooks on each decoder layer handle the transfer automatically:
+
+- **Forward pass:** Before a layer executes, its frozen params are loaded to GPU. The next layer is
+  prefetched asynchronously on a separate CUDA stream for overlap.
+- **Backward pass:** Same pattern in reverse — the current layer's frozen params are loaded and the
+  previous layer is prefetched.
+
+After each layer finishes, its frozen params are offloaded back to CPU pinned memory.
+
+This approach trades some CPU-GPU transfer overhead for significant GPU memory savings — the freed memory
+is roughly equal to the size of all frozen parameters across all decoder layers, minus one layer's worth
+that is kept on GPU at any given time.
+
+**Requirements:**
+
+- CUDA GPU (CPU-only training is not supported for this feature)
+- Works with any HuggingFace model architecture that uses decoder layers (Llama, Mistral, Qwen, etc.)
+- Best combined with LoRA/QLoRA where most parameters are frozen

docs/multimodal.qmd

@@ -20,6 +20,7 @@ format:
 - [Gemma-3n](#sec-gemma-3n)
 - [Qwen2-VL](#sec-qwen2-vl)
 - [Qwen2.5-VL](#sec-qwen25-vl)
+- [Qwen3.5](#sec-qwen3-5)
 - [GLM-4.6V](#sec-glm-4-6v)
 - [SmolVLM2](#sec-smolvlm2)
 - [LFM2-VL](#sec-lfm2-vl)
@@ -191,6 +192,14 @@ base_model: Qwen/Qwen3-VL-4B-Instruct
 chat_template: qwen2_vl  # same as qwen2-vl
 ```
 
+### Qwen3.5 {#sec-qwen3-5}
+
+```yaml
+base_model: Qwen/Qwen3.5-9B
+
+chat_template: qwen3_5
+```
+
 ### GLM-4.6V {#sec-glm-4-6v}
 
 Both GLM-4.6V (106B MoE) and GLM-4.6V-Flash (9B) are supported.

docs/optimizations.qmd

@@ -54,6 +54,13 @@ These techniques save VRAM by changing how activations are handled.
 - Activation Offloading: moves activations to CPU RAM or disk, trading I/O overhead for VRAM.
 - Learn more: [Gradient Checkpointing and Offloading Docs](gradient_checkpointing.qmd)
 
+### Layer Offloading
+
+Offloads frozen (non-trainable) decoder layer parameters to CPU and streams them back to GPU one layer at a time during forward/backward passes using CUDA stream prefetching. Especially effective for LoRA/QLoRA where most parameters are frozen.
+
+- **Config:** `layer_offloading: true`
+- **Learn more:** [Layer Offloading Docs](gradient_checkpointing.qmd#enabling-layer-offloading)
+
 ### Cut Cross Entropy (CCE)
 
 Reduces VRAM usage by using an optimized cross-entropy loss calculation.

examples/colab-notebooks/colab-axolotl-example.ipynb

@@ -40,7 +40,7 @@
     "%%capture\n",
     "# This step can take ~5-10 minutes to install dependencies\n",
     "!pip install --no-build-isolation axolotl[flash-attn]>=0.9.1\n",
-    "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@fa9a7fe\""
+    "!pip install \"cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@63b15e6\""
    ]
   },
   {

examples/gemma3/gemma-3-1b-qlora.yml

@@ -1,8 +1,5 @@
 base_model: google/gemma-3-1b-it
 
-model_type: Gemma3ForCausalLM
-cls_model_config: Gemma3TextConfig
-
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
@@ -27,6 +24,11 @@ datasets:
 val_set_size: 0.0
 output_dir: ./outputs/out
 
+# Freeze vision tower
+unfrozen_parameters:
+  - ^model\.language_model\..*
+  - ^lm_head\..*
+
 adapter: qlora
 lora_r: 32
 lora_alpha: 16

examples/gemma3/gemma-3-270m-qlora.yml

@@ -1,8 +1,5 @@
 base_model: google/gemma-3-270m-it
 
-model_type: Gemma3ForCausalLM
-cls_model_config: Gemma3TextConfig
-
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 
@@ -27,6 +24,11 @@ datasets:
 val_set_size: 0.0
 output_dir: ./outputs/out
 
+# Freeze vision tower
+unfrozen_parameters:
+  - ^model\.language_model\..*
+  - ^lm_head\..*
+
 adapter: qlora
 lora_r: 32
 lora_alpha: 16

examples/gemma3/gemma-3-4b-qlora.yml

@@ -1,9 +1,5 @@
 base_model: google/gemma-3-4b-it
 
-# Need to set else transformers tries to load vision too
-model_type: Gemma3ForCausalLM
-cls_model_config: Gemma3TextConfig
-
 load_in_4bit: true
 
 # gemma3 doesn't seem to play nice with ddp
@@ -24,6 +20,11 @@ dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./outputs/out
 
+# Freeze vision tower
+unfrozen_parameters:
+  - ^model\.language_model\..*
+  - ^lm_head\..*
+
 adapter: qlora
 lora_model_dir:
 

examples/mistral4/README.md

@@ -6,9 +6,6 @@ Thanks to the team at MistralAI for giving us early access to prepare for this r
 
 ## Getting started
 
-Note: Training this model requires weights in BF16 which we will link to later.
-Users interested in training can convert / descale the existing FP8 weights.
-
 1. Install Axolotl following the [installation guide](https://docs.axolotl.ai/docs/installation.html).
 
 2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage

examples/mistral4/fft-text.yml

@@ -1,4 +1,4 @@
-base_model: mistralai/Mistral-Small-4-119B-2603
+base_model: axolotl-ai-co/Mistral-Small-4-119B-2603-BF16
 
 plugins:
   - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

examples/mistral4/fft-vision.yml

@@ -1,4 +1,4 @@
-base_model: mistralai/Mistral-Small-4-119B-2603
+base_model: axolotl-ai-co/Mistral-Small-4-119B-2603-BF16
 processor_type: AutoProcessor
 
 plugins:

examples/mistral4/qlora-text.yml

@@ -1,4 +1,4 @@
-base_model: mistralai/Mistral-Small-4-119B-2603
+base_model: axolotl-ai-co/Mistral-Small-4-119B-2603-BF16
 
 plugins:
   - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

examples/mistral4/qlora-vision.yml

@@ -1,4 +1,4 @@
-base_model: mistralai/Mistral-Small-4-119B-2603
+base_model: axolotl-ai-co/Mistral-Small-4-119B-2603-BF16
 processor_type: AutoProcessor
 
 plugins:

examples/nemotron/nemotron-mini-4b-qlora.yaml

@@ -0,0 +1,57 @@
+base_model: nvidia/Nemotron-Mini-4B-Instruct
+
+load_in_8bit: false
+load_in_4bit: true
+
+datasets:
+  - path: fozziethebeat/alpaca_messages_2k_test
+    type: chat_template
+
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.1
+output_dir: ./outputs/nemotron-mini-4b-qlora
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_target_modules:
+  - q_proj
+  - k_proj
+  - v_proj
+  - o_proj
+  - up_proj
+  - down_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 1
+saves_per_epoch: 1
+
+special_tokens:

examples/qwen3.5/122b-a10b-moe-qlora-fsdp.yaml

@@ -0,0 +1,84 @@
+base_model: Qwen/Qwen3.5-122B-A10B
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+strict: false
+
+chat_template: qwen3_5
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_property_mappings:
+      role: from
+      content: value
+val_set_size: 0.0
+output_dir: ./outputs/out
+dataset_prepared_path: last_run_prepared
+
+sequence_len: 2048
+sample_packing: true
+
+load_in_4bit: true
+quantize_moe_experts: true
+adapter: qlora
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0
+lora_target_modules:
+  - q_proj
+  - k_proj
+  - v_proj
+  - o_proj
+# Regex matching to target shared experts too
+# lora_target_modules: 'model\.(language_model\.)?layers\.[\d]+\.(mlp|self_attn)\.(shared_expert\.)?(up|down|gate|gate_up|q|k|v|o)_proj'
+
+# Target experts
+# lora_target_parameters:
+#   - mlp.experts.gate_up_proj
+#   - mlp.experts.down_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch_4bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+lora_mlp_kernel: false
+lora_qkv_kernel: false
+lora_o_kernel: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:
+
+fsdp_config:
+  fsdp_version: 2
+  offload_params: true
+  cpu_ram_efficient_loading: false
+  auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  transformer_layer_cls_to_wrap: Qwen3_5MoeDecoderLayer
+  state_dict_type: FULL_STATE_DICT
+  sharding_strategy: FULL_SHARD
+  reshard_after_forward: true
+  activation_checkpointing: true

examples/qwen3.5/122b-a10b-moe-qlora.yaml

@@ -32,7 +32,11 @@ lora_target_modules:
   - v_proj
   - o_proj
 
-#lora_target_parameters:
+# Regex matching to target shared experts too
+# lora_target_modules: 'model\.(language_model\.)?layers\.[\d]+\.(mlp|self_attn)\.(shared_expert\.)?(up|down|gate|gate_up|q|k|v|o)_proj'
+
+# Target experts
+# lora_target_parameters:
 #   - mlp.experts.gate_up_proj
 #   - mlp.experts.down_proj
 
@@ -52,7 +56,6 @@ learning_rate: 0.0002
 bf16: auto
 tf32: true
 
-
 lora_mlp_kernel: false
 lora_qkv_kernel: false
 lora_o_kernel: false

examples/qwen3.5/27b-fft.yaml

@@ -0,0 +1,59 @@
+base_model: Qwen/Qwen3.5-27B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+# Full fine-tune (FFT) of the text-only path of Qwen3.5-27B.
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+strict: false
+
+chat_template: qwen3_5
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_property_mappings:
+      role: from
+      content: value
+val_set_size: 0.0
+output_dir: ./outputs/out
+dataset_prepared_path: last_run_prepared
+
+sequence_len: 2048
+sample_packing: true
+
+# Freeze vision encoder
+unfrozen_parameters:
+  - model\.language_model\..*
+  - lm_head\..*
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:

examples/qwen3.5/27b-qlora-fsdp.yaml

@@ -0,0 +1,81 @@
+base_model: Qwen/Qwen3.5-27B
+
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+strict: false
+
+chat_template: qwen3_5
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_property_mappings:
+      role: from
+      content: value
+val_set_size: 0.0
+output_dir: ./outputs/out
+dataset_prepared_path: last_run_prepared
+
+sequence_len: 2048
+sample_packing: true
+
+load_in_4bit: true
+adapter: qlora
+lora_r: 16
+lora_alpha: 32
+lora_target_modules:
+  - q_proj
+  - k_proj
+  - v_proj
+  - o_proj
+  - down_proj
+  - up_proj
+  # Uncomment below to also target the linear attention projections.
+  # These use separate in_proj_qkv / in_proj_z / out_proj (Qwen3.5-specific).
+  # - linear_attn.in_proj_qkv
+  # - linear_attn.in_proj_z
+  # - linear_attn.out_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch_4bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:
+
+fsdp_config:
+  fsdp_version: 2
+  offload_params: false
+  cpu_ram_efficient_loading: false
+  auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  transformer_layer_cls_to_wrap: Qwen3_5DecoderLayer
+  state_dict_type: FULL_STATE_DICT
+  sharding_strategy: FULL_SHARD
+  reshard_after_forward: true
+  activation_checkpointing: true

examples/qwen3.5/27b-qlora.yaml

@@ -1,9 +1,7 @@
 base_model: Qwen/Qwen3.5-27B
+
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
-# Note: Qwen3.5 is an early-fusion VLM (image+text). This config fine-tunes
-# the text-only path. For multimodal (image+text) fine-tuning, add image
-# columns to your dataset following axolotl's multimodal dataset format.
 
 plugins:
   - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

examples/qwen3.5/35b-a3b-moe-qlora-fsdp.yaml

@@ -0,0 +1,85 @@
+base_model: Qwen/Qwen3.5-35B-A3B
+
+plugins:
+  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
+strict: false
+
+chat_template: qwen3_5
+datasets:
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_property_mappings:
+      role: from
+      content: value
+val_set_size: 0.0
+output_dir: ./outputs/out
+dataset_prepared_path: last_run_prepared
+
+sequence_len: 2048
+sample_packing: true
+
+load_in_4bit: true
+quantize_moe_experts: true
+adapter: qlora
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0
+lora_target_modules:
+  - q_proj
+  - k_proj
+  - v_proj
+  - o_proj
+
+# Regex matching to target shared experts too
+# lora_target_modules: 'model\.(language_model\.)?layers\.[\d]+\.(mlp|self_attn)\.(shared_expert\.)?(up|down|gate|gate_up|q|k|v|o)_proj'
+
+# Target experts
+# lora_target_parameters:
+#   - mlp.experts.gate_up_proj
+#   - mlp.experts.down_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch_4bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+lora_mlp_kernel: false
+lora_qkv_kernel: false
+lora_o_kernel: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:
+
+fsdp_config:
+  fsdp_version: 2
+  offload_params: true
+  cpu_ram_efficient_loading: false
+  auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  transformer_layer_cls_to_wrap: Qwen3_5MoeDecoderLayer
+  state_dict_type: FULL_STATE_DICT
+  sharding_strategy: FULL_SHARD
+  reshard_after_forward: true
+  activation_checkpointing: true

examples/qwen3.5/35b-a3b-moe-qlora.yaml

@@ -32,7 +32,11 @@ lora_target_modules:
   - v_proj
   - o_proj
 
-#lora_target_parameters:
+# Regex matching to target shared experts too
+# lora_target_modules: 'model\.(language_model\.)?layers\.[\d]+\.(mlp|self_attn)\.(shared_expert\.)?(up|down|gate|gate_up|q|k|v|o)_proj'
+
+# Target experts
+# lora_target_parameters:
 #   - mlp.experts.gate_up_proj
 #   - mlp.experts.down_proj
 

examples/qwen3.5/9b-fft-vision.yaml

@@ -0,0 +1,49 @@
+base_model: Qwen/Qwen3.5-9B
+processor_type: AutoProcessor
+
+# Required for multimodal training
+skip_prepare_dataset: true
+remove_unused_columns: false
+sample_packing: false
+
+chat_template: qwen3_5
+datasets:
+  - path: HuggingFaceH4/llava-instruct-mix-vsft
+    type: chat_template
+    split: train[:1%]
+
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.0
+output_dir: ./outputs/out
+
+sequence_len: 4096
+pad_to_sequence_len: false
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+warmup_ratio: 0.1
+evals_per_epoch: 1
+saves_per_epoch: 1
+weight_decay: 0.0
+special_tokens:

examples/qwen3.5/9b-lora-vision.yaml

@@ -1,10 +1,6 @@
-base_model: Qwen/Qwen3.5-7B
+base_model: Qwen/Qwen3.5-9B
 processor_type: AutoProcessor
 
-# Qwen3.5-7B and above are early-fusion VLMs (Qwen3_5ForConditionalGeneration).
-# Vision and text tokens are processed together by the same transformer layers.
-# Note: Qwen3.5-2B is a text-only model — the smallest VLM is Qwen3.5-7B.
-
 # These 3 lines are required for vision/multimodal training
 skip_prepare_dataset: true
 remove_unused_columns: false
@@ -30,8 +26,6 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 # Targets the language model attention and MLP layers.
-# Qwen3.5 is early-fusion: all layers (including those seeing vision tokens) share
-# the same transformer stack, so standard attention targets work for both modalities.
 lora_target_modules:
   - q_proj
   - k_proj

examples/qwen3.5/README.md

@@ -1,15 +1,6 @@
 # Finetune Qwen3.5 with Axolotl
 
-[Qwen3.5](https://huggingface.co/collections/Qwen/qwen35-68452f3bc6e4b7cfb4e1c803) is a hybrid architecture model series combining Gated DeltaNet linear attention with standard Transformer attention. Models from 7B onwards are early-fusion vision-language models (`Qwen3_5ForConditionalGeneration`), meaning vision and text tokens are processed through the same transformer stack. The 2B variant is text-only.
-
-Available configs:
-
-| Config | Model | Type |
-|---|---|---|
-| `27b-qlora.yaml` | Qwen3.5-27B | Dense VLM, text-only path |
-| `35b-a3b-moe-qlora.yaml` | Qwen3.5-35B-A3B | MoE, text-only path |
-| `122b-a10b-moe-qlora.yaml` | Qwen3.5-122B-A10B | MoE, text-only path |
-| `7b-lora-vision.yaml` | Qwen3.5-7B | Vision+text (multimodal) |
+[Qwen3.5](https://huggingface.co/collections/Qwen/qwen35) is a hybrid architecture model series combining Gated DeltaNet linear attention with standard Transformer attention. All Qwen3.5 models are early-fusion vision-language models: dense variants use `Qwen3_5ForConditionalGeneration` and MoE variants use `Qwen3_5MoeForConditionalGeneration`.
 
 ## Getting started
 
@@ -18,35 +9,69 @@ Available configs:
 2. Install [Cut Cross Entropy](https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy) to reduce training VRAM usage.
 
 3. Install FLA for sample packing support with the Gated DeltaNet linear attention layers:
-```bash
-pip3 uninstall -y causal-conv1d && pip3 install flash-linear-attention==0.4.1
+  ```bash
+  pip3 uninstall -y causal-conv1d && pip3 install flash-linear-attention==0.4.1
+  ```
+  > FLA is required when `sample_packing: true`. Without it, training raises a `RuntimeError` on packed sequences. Vision configs use `sample_packing: false` so FLA is optional there.
+
+4. Pick any config from the table below and run:
+
+    ```bash
+    axolotl train examples/qwen3.5/<config>.yaml
+    ```
+
+Available configs:
+
+| Config | Model | Type | Peak VRAM |
+|---|---|---|---|
+| `9b-lora-vision.yaml` | Qwen3.5-9B | Vision+text LoRA, single GPU | — |
+| `9b-fft-vision.yaml` | Qwen3.5-9B | Vision+text FFT, single GPU | ~61 GiB |
+| `27b-qlora.yaml` | Qwen3.5-27B | Dense, text-only QLoRA | ~47 GiB |
+| `27b-fft.yaml` | Qwen3.5-27B | Dense, text-only FFT (vision frozen) | ~53 GiB |
+| `27b-qlora-fsdp.yaml` | Qwen3.5-27B | Dense, text-only QLoRA + FSDP2 | — |
+| `35b-a3b-moe-qlora.yaml` | Qwen3.5-35B-A3B | MoE, text-only QLoRA | — |
+| `35b-a3b-moe-qlora-fsdp.yaml` | Qwen3.5-35B-A3B | MoE, text-only QLoRA + FSDP2 | — |
+| `122b-a10b-moe-qlora.yaml` | Qwen3.5-122B-A10B | MoE, text-only QLoRA | — |
+| `122b-a10b-moe-qlora-fsdp.yaml` | Qwen3.5-122B-A10B | MoE, text-only QLoRA + FSDP2 | — |
+
+### Gated DeltaNet Linear Attention
+
+Qwen3.5 interleaves standard attention with Gated DeltaNet linear attention layers. To apply LoRA to them, add to `lora_target_modules`:
+
+```yaml
+lora_target_modules:
+  # ... standard projections ...
+  - linear_attn.in_proj_qkv
+  - linear_attn.in_proj_z
+  - linear_attn.out_proj
 ```
-> FLA is required when `sample_packing: true`. Without it, training raises a `RuntimeError` on packed sequences. Vision configs use `sample_packing: false` so FLA is optional there.
 
-4. Run a finetuning example:
+### Routed Experts (MoE)
 
-```bash
-# Dense 27B text-only (QLoRA, ~47 GiB VRAM with sample packing)
-axolotl train examples/qwen3.5/27b-qlora.yaml
+To apply LoRA to routed expert parameters, add `lora_target_parameters`:
 
-# MoE 35B-A3B text-only (QLoRA)
-axolotl train examples/qwen3.5/35b-a3b-moe-qlora.yaml
+```yaml
+lora_target_parameters:
+  - mlp.experts.gate_up_proj
+  - mlp.experts.down_proj
+#  - mlp.gate.weight  # router
+```
 
-# MoE 122B-A10B text-only (QLoRA)
-axolotl train examples/qwen3.5/122b-a10b-moe-qlora.yaml
+### Shared Experts (MoE)
 
-# 7B vision+text (LoRA, multimodal dataset)
-axolotl train examples/qwen3.5/7b-lora-vision.yaml
+Routed experts and shared experts both have `gate_up_proj`/`down_proj`, so a plain module name in `lora_target_modules` would match both. Use a regex to target only attention and shared expert projections, while `lora_target_parameters` above handles routed experts separately:
+
+```yaml
+lora_target_modules: 'model\.(language_model\.)?layers\.[\d]+\.(mlp|self_attn)\.(shared_expert\.)?(up|down|gate|gate_up|q|k|v|o)_proj'
 ```
 
 ### TIPS
 
-- For inference, you can experiment with `temperature: 0.7`, `top_p: 0.8`, `top_k: 20`, and `min_p: 0`.
-- You can run a full finetuning by removing `adapter: qlora` and `load_in_4bit: true`. See [Multi-GPU](#optimization-guides) below.
+- For inference hyp, please see the respective model card details.
+- You can run a full finetuning of smaller configs by removing `adapter: qlora` and `load_in_4bit: true`. See [Multi-GPU](#optimization-guides) below.
 - Read more on loading your own dataset at [docs](https://docs.axolotl.ai/docs/dataset_loading.html).
 - The dataset format follows the OpenAI Messages format as seen [here](https://docs.axolotl.ai/docs/dataset-formats/conversation.html#chat_template).
-- For **multimodal** finetuning, set `processor_type: AutoProcessor`, `skip_prepare_dataset: true`, and `remove_unused_columns: false` as shown in `7b-lora-vision.yaml`.
-- The Gated DeltaNet linear attention layers (`linear_attn.*`) can optionally be added to `lora_target_modules` — they are commented out by default.
+- For **multimodal** finetuning, set `processor_type: AutoProcessor`, `skip_prepare_dataset: true`, and `remove_unused_columns: false` as shown in `9b-lora-vision.yaml`.
 
 ## Optimization Guides
 

pyproject.toml

@@ -61,5 +61,11 @@ skip-magic-trailing-comma = false
 line-ending = "auto"
 docstring-code-format = false
 
+[tool.pytest.ini_options]
+addopts = "-m 'not slow'"
+markers = [
+    "slow: marks tests as slow",
+]
+
 [tool.uv.extra-build-dependencies]
 axolotl = ["huggingface_hub"]

scripts/cutcrossentropy_install.py

@@ -29,5 +29,5 @@ UV_PREFIX = "uv " if USE_UV else ""
 
 print(
     UNINSTALL_PREFIX
-    + f'{UV_PREFIX}pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@fa9a7fe"'
+    + f'{UV_PREFIX}pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@63b15e6"'
 )

setup.py

@@ -81,16 +81,23 @@ def parse_requirements(extras_require_map):
                     f"https://download.pytorch.org/whl/{torch_cuda_version}"
                 )
 
-            if (major, minor) >= (2, 9):
+            if (major, minor) >= (2, 10):
+                extras_require_map.pop("fbgemm-gpu")
+                extras_require_map["fbgemm-gpu"] = [
+                    "fbgemm-gpu==1.5.0",
+                    "fbgemm-gpu-genai==1.5.0",
+                ]
+                if not install_xformers:
+                    _install_requires.pop(_install_requires.index(xformers_version))
+                extras_require_map["vllm"] = ["vllm==0.17.1"]
+            elif (major, minor) >= (2, 9):
                 extras_require_map.pop("fbgemm-gpu")
                 extras_require_map["fbgemm-gpu"] = [
                     "fbgemm-gpu==1.4.0",
                     "fbgemm-gpu-genai==1.4.2",
                 ]
-                extras_require_map["vllm"] = ["vllm==0.11.1"]
                 if not install_xformers:
                     _install_requires.pop(_install_requires.index(xformers_version))
-                extras_require_map["vllm"] = ["vllm==0.13.0"]
                 if patch == 0:
                     extras_require_map["vllm"] = ["vllm==0.13.0"]
                 else:

src/axolotl/cli/checks.py

@@ -3,6 +3,7 @@
 import os
 from pathlib import Path
 
+import httpcore
 from accelerate.commands.config import config_args
 from huggingface_hub import HfApi
 from huggingface_hub.utils import LocalTokenNotFoundError
@@ -47,7 +48,7 @@ def check_user_token() -> bool:
             "Error verifying HuggingFace token. Remember to log in using `hf auth login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets."
         )
         return False
-    except HTTPError:
+    except (HTTPError, httpcore.ConnectError):
         LOG.warning(
             "Error accessing HuggingFace. This may be due to a network issue or rate limiting."
         )

src/axolotl/core/builders/base.py

@@ -353,6 +353,30 @@ class TrainerBuilderBase(abc.ABC):
                 adam_kwargs["eps"] = (eps1, eps2)
 
                 optimizer_kwargs.update(adam_kwargs)
+            elif self.cfg.optimizer == "flash_adamw":
+                from flashoptim import FlashAdamW
+
+                optimizer_cls = FlashAdamW
+                optimizer_kwargs.update(adam_kwargs)
+            elif self.cfg.optimizer == "flash_adam":
+                from flashoptim import FlashAdam
+
+                optimizer_cls = FlashAdam
+                optimizer_kwargs.update(adam_kwargs)
+            elif self.cfg.optimizer == "flash_sgd":
+                from flashoptim import FlashSGD
+
+                optimizer_cls = FlashSGD
+            elif self.cfg.optimizer == "flash_sgdw":
+                from flashoptim import FlashSGDW
+
+                optimizer_cls = FlashSGDW
+            elif self.cfg.optimizer == "flash_lion":
+                from flashoptim import FlashLion
+
+                optimizer_cls = FlashLion
+                if "betas" in adam_kwargs:
+                    optimizer_kwargs["betas"] = adam_kwargs["betas"]
             else:
                 raise ValueError(
                     f"Unhandled optimizer: {self.cfg.optimizer}. Please raise an Issue."
@@ -484,6 +508,8 @@ class TrainerBuilderBase(abc.ABC):
             training_args_kwargs["accelerator_config"] = AcceleratorConfig()
 
     def _configure_gradient_checkpointing(self, training_args_kwargs: dict):
+        if self.cfg.layer_offloading:
+            training_args_kwargs["layer_offloading"] = True
         if self.cfg.activation_offloading is True:
             # don't use the HF gradient checkpointing, manually wrap
             training_args_kwargs["gradient_checkpointing"] = False

src/axolotl/core/builders/causal.py

@@ -421,6 +421,13 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
             trainer_kwargs["dataset_tags"] = [
                 d["path"] for d in self.cfg.datasets if not Path(d["path"]).is_dir()
             ]
+        # TRL's RewardTrainer validates num_labels=1 on pre-loaded models; ensure the
+        # config reflects this regardless of how the model was instantiated.
+        if (
+            self.cfg.reward_model
+            and getattr(self.model.config, "num_labels", None) != 1
+        ):
+            self.model.config.num_labels = 1
         trainer = trainer_cls(
             model=self.model,
             train_dataset=self.train_dataset,

src/axolotl/core/builders/rl.py

@@ -208,7 +208,11 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
 
         if self.eval_dataset:
             trainer_kwargs["eval_dataset"] = self.eval_dataset
-        if self.cfg.adapter and self.peft_config and self.cfg.rl is not RLType.GRPO:
+        if (
+            self.cfg.adapter
+            and self.peft_config
+            and self.cfg.rl not in (RLType.GRPO, RLType.ORPO)
+        ):
             trainer_kwargs["peft_config"] = self.peft_config
         if self.cfg.precompute_ref_log_probs is not None:
             trainer_kwargs["precompute_ref_log_probs"] = (

src/axolotl/core/trainers/base.py

@@ -29,10 +29,12 @@ from transformers.utils import SAFE_WEIGHTS_NAME, is_peft_available
 from trl.experimental.utils import pad_to_length
 from typing_extensions import override
 
+from axolotl.core.trainers.constants import TOKENS_STATE_FILE
 from axolotl.core.trainers.mixins import (
     ActivationOffloadingMixin,
     CheckpointSaveMixin,
     DistributedParallelMixin,
+    LayerOffloadingMixin,
     OptimizerMixin,
     PackingMixin,
     RngLoaderMixin,
@@ -51,8 +53,6 @@ from axolotl.utils.samplers import MultipackBatchSampler, get_dataset_lengths
 
 LOG = get_logger(__name__)
 
-TOKENS_STATE_FILE = "tokens_state."
-
 REDUCTION_FNS = {
     "mean": torch.mean,
     "min": torch.min,
@@ -67,6 +67,7 @@ class AxolotlTrainer(
     OptimizerMixin,
     RngLoaderMixin,
     CheckpointSaveMixin,
+    LayerOffloadingMixin,
     ActivationOffloadingMixin,
     DistributedParallelMixin,
     Trainer,

src/axolotl/core/trainers/constants.py

@@ -0,0 +1 @@
+TOKENS_STATE_FILE = "tokens_state.json"

src/axolotl/core/trainers/dpo/args.py

@@ -2,7 +2,8 @@
 Axolotl specific DPO args
 """
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
+from typing import Optional
 
 from trl import DPOConfig
 
@@ -16,3 +17,4 @@ class AxolotlDPOConfig(AxolotlTrainingMixins, DPOConfig):
     """
 
     dpo_norm_loss: bool | None = False
+    rpo_alpha: Optional[float] = field(default=None)

src/axolotl/core/trainers/mixins/__init__.py

@@ -4,6 +4,7 @@
 
 from .activation_checkpointing import ActivationOffloadingMixin
 from .checkpoints import CheckpointSaveMixin
+from .layer_offloading import LayerOffloadingMixin
 from .distributed_parallel import DistributedParallelMixin
 from .optimizer import OptimizerMixin
 from .packing import PackingMixin

src/axolotl/core/trainers/mixins/layer_offloading.py

@@ -0,0 +1,304 @@
+"""
+Trainer mixin for layer-wise parameter offloading to CPU.
+
+Offloads frozen (non-trainable) parameters in decoder layers to CPU, then uses
+forward/backward hooks to stream them on/off GPU one layer at a time with CUDA
+stream prefetching. Trainable parameters (e.g. LoRA weights) stay on GPU always.
+
+Forward:  pre-hook loads layer N's frozen params to GPU (prefetches N+1 on
+          transfer stream), post-hook offloads layer N-1's frozen params.
+Backward: same in reverse order.
+"""
+
+import contextlib
+
+import torch
+import torch.nn as nn
+from transformers import Trainer
+
+from axolotl.utils.logging import get_logger
+
+LOG = get_logger(__name__)
+
+
+def _find_decoder_layers(model: nn.Module) -> tuple[nn.ModuleList | None, list[str]]:
+    """Recursively search the model for the decoder layer ModuleList.
+
+    Finds any ModuleList whose children have 'DecoderLayer' in their class name.
+    Handles all common HF architectures including VLM wrappers (e.g. Qwen3.5-MoE
+    where layers are at model.language_model.layers).
+    """
+    # BFS to find the first ModuleList containing decoder layers
+    queue = [model]
+    while queue:
+        m = queue.pop(0)
+        for _name, child in m.named_children():
+            if isinstance(child, nn.ModuleList) and len(child) > 0:
+                first_type = type(child[0]).__name__
+                if "DecoderLayer" in first_type or "TransformerBlock" in first_type:
+                    layer_types = list({type(layer).__name__ for layer in child})
+                    return child, layer_types
+            else:
+                queue.append(child)
+
+    return None, []
+
+
+def _get_frozen_params(layer: nn.Module) -> list[tuple[str, nn.Parameter]]:
+    """Get all non-trainable parameters in a layer."""
+    return [(n, p) for n, p in layer.named_parameters() if not p.requires_grad]
+
+
+class LayerOffloadManager:
+    """Manages offloading frozen decoder layer params to CPU and streaming
+    them back during forward/backward with CUDA stream overlap.
+
+    Only frozen (requires_grad=False) parameters are offloaded.
+    Trainable parameters (LoRA weights, etc.) remain on GPU at all times.
+    """
+
+    def __init__(
+        self,
+        model: nn.Module,
+        num_prefetch: int = 1,
+    ):
+        self.model = model
+        self.num_prefetch = num_prefetch
+        self._hooks: list = []
+        self._device = None
+
+        # Find decoder layers
+        self.layers, layer_types = _find_decoder_layers(model)
+        if self.layers is None:
+            LOG.warning(
+                "LayerOffloadManager: no decoder layers found, offloading disabled"
+            )
+            self.enabled = False
+            return
+
+        self.enabled = True
+        self.n_layers = len(self.layers)
+        LOG.info(
+            f"Layer offloading: found {self.n_layers} layers ({', '.join(layer_types)})"
+        )
+
+        # Determine GPU device
+        for p in model.parameters():
+            if p.device.type == "cuda":
+                self._device = p.device
+                break
+        if self._device is None:
+            LOG.warning("LayerOffloadManager: no CUDA parameters found")
+            self.enabled = False
+            return
+
+        # Transfer stream for async prefetch
+        self._transfer_stream = torch.cuda.Stream(device=self._device)
+
+        # Track which layers have their frozen params on GPU
+        self._on_gpu: set[int] = set(range(self.n_layers))
+
+        # Cache: frozen param references per layer (list of (name, param) tuples)
+        self._frozen_params: list[list[tuple[str, nn.Parameter]]] = [
+            _get_frozen_params(self.layers[i]) for i in range(self.n_layers)
+        ]
+
+        # CPU storage: pinned tensors for each layer's frozen params
+        # Populated on first offload
+        self._cpu_data: list[dict[str, torch.Tensor]] = [
+            {} for _ in range(self.n_layers)
+        ]
+
+        # Offload all layers upfront
+        self._offload_all()
+
+        # Release cached memory blocks back to the driver
+        torch.cuda.empty_cache()
+
+    def _offload_all(self):
+        """Move all frozen params in all decoder layers to CPU."""
+        mem_before = torch.cuda.memory_allocated(self._device)
+        for i in range(self.n_layers):
+            self._offload_layer(i)
+        mem_after = torch.cuda.memory_allocated(self._device)
+        freed = (mem_before - mem_after) / 1e6
+        LOG.info(
+            f"Layer offloading: offloaded frozen params from {self.n_layers} layers, "
+            f"freed {freed:.0f} MB GPU memory"
+        )
+
+    def _offload_layer(self, idx: int):
+        """Move frozen params of layer idx to CPU pinned memory."""
+        if idx not in self._on_gpu:
+            return
+        for name, param in self._frozen_params[idx]:
+            if param.device.type != "cuda":
+                continue
+            # Allocate pinned CPU tensor on first offload
+            if name not in self._cpu_data[idx]:
+                self._cpu_data[idx][name] = torch.empty_like(
+                    param.data, device="cpu", pin_memory=True
+                )
+            cpu_buf = self._cpu_data[idx][name]
+            # Async copy GPU -> CPU (on transfer stream for overlap)
+            cpu_buf.copy_(param.data, non_blocking=True)
+            # Point parameter at a dummy CPU tensor to free GPU memory
+            param.data = cpu_buf
+        self._on_gpu.discard(idx)
+
+    def _load_layer(self, idx: int, stream=None):
+        """Move frozen params of layer idx back to GPU."""
+        if idx in self._on_gpu or idx < 0 or idx >= self.n_layers:
+            return
+        ctx = (
+            torch.cuda.stream(stream)
+            if stream is not None
+            else contextlib.nullcontext()
+        )
+        with ctx:
+            for _name, param in self._frozen_params[idx]:
+                if param.device.type == "cuda":
+                    continue
+                gpu_data = param.data.to(self._device, non_blocking=True)
+                param.data = gpu_data
+        self._on_gpu.add(idx)
+
+    def _prefetch_layer(self, idx: int):
+        """Async prefetch layer idx on the transfer stream."""
+        if idx in self._on_gpu or idx < 0 or idx >= self.n_layers:
+            return
+        self._transfer_stream.wait_stream(torch.cuda.default_stream(self._device))
+        self._load_layer(idx, stream=self._transfer_stream)
+
+    def _wait_transfer(self):
+        """Make default stream wait for any in-flight transfers."""
+        torch.cuda.default_stream(self._device).wait_stream(self._transfer_stream)
+
+    def setup_hooks(self):
+        """Register forward and backward hooks on each decoder layer."""
+        if not self.enabled:
+            return
+
+        for idx in range(self.n_layers):
+            layer = self.layers[idx]
+
+            def make_pre_fwd(i):
+                def hook(module, args):
+                    # Ensure this layer is on GPU
+                    if i not in self._on_gpu:
+                        self._load_layer(i)
+                    self._wait_transfer()
+                    # Prefetch next layer(s)
+                    for offset in range(1, self.num_prefetch + 1):
+                        self._prefetch_layer(i + offset)
+
+                return hook
+
+            def make_post_fwd(i):
+                def hook(module, args, output):
+                    # Offload previous layer (no longer needed in forward)
+                    if i > 0:
+                        self._offload_layer(i - 1)
+                    # Offload last layer after forward
+                    if i == self.n_layers - 1:
+                        self._offload_layer(i)
+
+                return hook
+
+            def make_pre_bwd(i):
+                def hook(module, grad_output):
+                    # Load this layer for backward
+                    if i not in self._on_gpu:
+                        self._load_layer(i)
+                    self._wait_transfer()
+                    # Prefetch previous layer(s)
+                    for offset in range(1, self.num_prefetch + 1):
+                        self._prefetch_layer(i - offset)
+
+                return hook
+
+            def make_post_bwd(i):
+                def hook(module, grad_input, grad_output):
+                    # Offload the layer above
+                    if i < self.n_layers - 1:
+                        self._offload_layer(i + 1)
+                    # Offload first layer after backward
+                    if i == 0:
+                        self._offload_layer(i)
+
+                return hook
+
+            h1 = layer.register_forward_pre_hook(make_pre_fwd(idx))
+            h2 = layer.register_forward_hook(make_post_fwd(idx))
+            h3 = layer.register_full_backward_pre_hook(make_pre_bwd(idx))
+            h4 = layer.register_full_backward_hook(make_post_bwd(idx))
+            self._hooks.extend([h1, h2, h3, h4])
+
+    def remove_hooks(self):
+        """Remove all hooks and restore layers to GPU."""
+        for h in self._hooks:
+            h.remove()
+        self._hooks.clear()
+        if self.enabled:
+            for i in range(self.n_layers):
+                if i not in self._on_gpu:
+                    self._load_layer(i)
+
+    def pre_step(self):
+        """Called before each training step — ensure layers start offloaded."""
+        if not self.enabled:
+            return
+        for i in list(self._on_gpu):
+            self._offload_layer(i)
+        # Prefetch layer 0 for forward
+        self._prefetch_layer(0)
+
+    def post_step(self):
+        """Called after each training step — ensure layers are offloaded."""
+        if not self.enabled:
+            return
+        for i in list(self._on_gpu):
+            self._offload_layer(i)
+        # Prefetch layer 0 for next step
+        self._prefetch_layer(0)
+
+
+class _LayerOffloadContext:
+    """Context manager wrapping pre_step / post_step around a training step."""
+
+    def __init__(self, manager: LayerOffloadManager):
+        self.manager = manager
+
+    def __enter__(self):
+        self.manager.pre_step()
+        return self
+
+    def __exit__(self, *args):
+        self.manager.post_step()
+
+
+class LayerOffloadingMixin(Trainer):
+    """
+    Trainer mixin class for layer-wise parameter offloading to CPU.
+
+    Offloads frozen decoder layer params to CPU at init, then streams them
+    on/off GPU one layer at a time during each training step.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        if getattr(self.args, "layer_offloading", False):
+            LOG.info("Layer parameter offloading enabled")
+            self._layer_offload_manager = LayerOffloadManager(
+                model=self.model,
+                num_prefetch=1,
+            )
+            self._layer_offload_manager.setup_hooks()
+            self._layer_offload_ctx = _LayerOffloadContext(self._layer_offload_manager)
+        else:
+            self._layer_offload_manager = None
+            self._layer_offload_ctx = contextlib.nullcontext()
+
+    def training_step(self, *args, **kwargs):
+        with self._layer_offload_ctx:
+            return super().training_step(*args, **kwargs)

src/axolotl/core/training_args_base.py

@@ -235,6 +235,13 @@ class AxolotlTrainingMixins:
         metadata={"help": "Use activation offloading with CUDA streams for training."},
     )
 
+    layer_offloading: bool | None = field(
+        default=None,
+        metadata={
+            "help": "Offload model layer parameters to CPU during forward, prefetch back during backward."
+        },
+    )
+
     # multi-modal section
 
     image_size: int | tuple[int, int] | None = field(

src/axolotl/integrations/cut_cross_entropy/README.md

@@ -19,7 +19,7 @@ python scripts/cutcrossentropy_install.py | sh
 
 - If you are installing from pip
 ```bash
-pip3 uninstall -y cut-cross-entropy && pip3 install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@fa9a7fe"
+pip3 uninstall -y cut-cross-entropy && pip3 install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@63b15e6"
 ```
 
 ## Usage

src/axolotl/integrations/cut_cross_entropy/__init__.py

@@ -35,7 +35,7 @@ LOG = get_logger(__name__)
 
 _CCE_INSTALL_MESSAGE = (
     "Please install Axolotl's fork of cut_cross_entropy with transformers support using "
-    '`pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@fa9a7fe"`'
+    '`pip install "cut-cross-entropy[transformers] @ git+https://github.com/axolotl-ai-cloud/ml-cross-entropy.git@63b15e6"`'
 )
 
 

src/axolotl/integrations/kernels/constants.py

@@ -15,6 +15,7 @@ SPARSE_MOE_BLOCK = {
     "qwen2_moe": "Qwen2MoeSparseMoeBlock",
     "qwen3_moe": "Qwen3MoeSparseMoeBlock",
     "qwen3_5_moe": "Qwen3_5MoeSparseMoeBlock",
+    "qwen3_5_moe_text": "Qwen3_5MoeSparseMoeBlock",
     "qwen3_next": "Qwen3NextSparseMoeBlock",
     "qwen3_vl_moe": "Qwen3VLMoeTextSparseMoeBlock",
     # qwen3_omni_moe: Thinker (standard) + Talker (shared experts + shared_expert_gate)
@@ -58,7 +59,16 @@ def resolve_moe_block_classes(model_type: str):
 
     cls_names = entry if isinstance(entry, list) else [entry]
     module_path = f"transformers.models.{model_type}.modeling_{model_type}"
-    module = importlib.import_module(module_path)
+    try:
+        module = importlib.import_module(module_path)
+    except ModuleNotFoundError:
+        # Text sub-model types (e.g. qwen3_5_moe_text) share the parent module
+        if model_type.endswith("_text"):
+            parent_type = model_type.removesuffix("_text")
+            module_path = f"transformers.models.{parent_type}.modeling_{parent_type}"
+            module = importlib.import_module(module_path)
+        else:
+            raise
 
     classes = []
     for cls_name in cls_names:

src/axolotl/integrations/kernels/libs/scattermoe_lora/kernels/lora_ops.py

@@ -195,6 +195,36 @@ def _estimate_smem_usage(
 _SMEM_SLACK = 10_000
 
 
+def _estimate_register_pressure(
+    num_warps: int,
+    *tile_sizes: tuple[int, int],
+) -> float:
+    """Rough estimate of per-thread register footprint from live tile sizes.
+
+    This is a heuristic, NOT an accurate register count.  Triton uses tensor
+    core MMA fragments that pack multiple elements per register, and can spill
+    to local memory when the hardware limit (255 regs/thread) is exceeded.
+
+    The estimate is used to prune only truly extreme configs that would cause
+    excessive spilling or compilation failures.  The threshold is set high
+    (``_MAX_REGS_SOFT_LIMIT``) because the heuristic overestimates — it
+    doesn't account for MMA fragment packing.  Configs like M=64,N=64,K=64
+    (est ~520) work fine in practice via spilling.
+
+    Returns estimated registers per thread.
+    """
+    # Each thread in a warp holds ~1/32 of the tile elements
+    tile_regs = sum(r * c for r, c in tile_sizes) / 32
+    scalar_overhead = 40
+    return tile_regs + scalar_overhead
+
+
+# Soft limit for register pressure pruning.  Only prune configs with extreme
+# tile products (e.g. M=128,K=256,N=256) that reliably crash on Blackwell.
+# Moderate configs (M=64,N=64,K=64, est ~520) work via register spilling.
+_MAX_REGS_SOFT_LIMIT = 1024
+
+
 # =============================================================================
 # Forward Kernel: scatter2scatter with fused LoRA
 # =============================================================================
@@ -313,12 +343,11 @@ def _compute_expert_block_lora(
         B_blk_ptrs, mask=N_mask[:, None] & R_mask[None, :], other=0.0
     )  # [BLOCK_N, BLOCK_R]
 
-    # Cast xa_acc and b to same dtype for tl.dot (required when input is bf16/fp16)
-    # Both operands must match; cast to float32 (accumulator type) for precision.
-    b_f32 = b.to(tl.float32)
+    # tl.dot requires non-float32 inputs (tensor cores); cast back to input dtype
+    b_inp = b.to(INPUT_DTYPE)
 
     # (X @ A^T) @ B^T: [M, R] @ [R, N] -> [M, N]
-    lora_out = tl.dot(xa_acc, tl.trans(b_f32), allow_tf32=allow_tf32)
+    lora_out = tl.dot(xa_acc.to(INPUT_DTYPE), tl.trans(b_inp), allow_tf32=allow_tf32)
 
     acc += scaling * lora_out
     return acc
@@ -327,28 +356,29 @@ def _compute_expert_block_lora(
 def _scatter2scatter_lora_configs():
     """Generate forward kernel autotune configs.
 
-    Search space includes smaller tile sizes and fewer pipeline stages to
-    support GPUs with limited shared memory (e.g. ~99KB on some GPUs).
+    Search space includes BLOCK_M to allow trading token-tile size for
+    larger BLOCK_K/BLOCK_N tiles.  On GPUs with ~99KB SMEM, BLOCK_M=128
+    forces BLOCK_K=32 and BLOCK_N=32; BLOCK_M=64 allows BLOCK_K=128
+    (4× fewer inner-loop iterations).
 
     Search space:
-      BLOCK_N:    {32, 64, 128, 256}
+      BLOCK_M:    {32, 64, 128}
+      BLOCK_N:    {32, 64}
       BLOCK_K:    {32, 64, 128}
       num_warps:  {4, 8}
       num_stages: {3, 4, 5}
-
-    BLOCK_M is fixed at 128 (module-level constant, not autotuned in the
-    scatter2scatter pattern).
     """
     configs = []
-    for block_n, block_k, warps, stages in product(
-        [32, 64, 128, 256],  # BLOCK_N
+    for block_m, block_n, block_k, warps, stages in product(
+        [32, 64, 128],  # BLOCK_M
+        [32, 64],  # BLOCK_N
         [32, 64, 128],  # BLOCK_K
         [4, 8],  # num_warps
         [3, 4, 5],  # num_stages
     ):
         configs.append(
             triton.Config(
-                {"BLOCK_N": block_n, "BLOCK_K": block_k},
+                {"BLOCK_M": block_m, "BLOCK_N": block_n, "BLOCK_K": block_k},
                 num_stages=stages,
                 num_warps=warps,
             )
@@ -357,7 +387,7 @@ def _scatter2scatter_lora_configs():
 
 
 def _prune_fwd_configs(configs, named_args, **kwargs):
-    """Prune forward configs based on SMEM capacity.
+    """Prune forward configs based on SMEM capacity and register pressure.
 
     The forward kernel inner loop loads three tiles per pipeline stage:
       X[BLOCK_M, BLOCK_K], W[BLOCK_K, BLOCK_N], A[BLOCK_R, BLOCK_K].
@@ -373,23 +403,49 @@ def _prune_fwd_configs(configs, named_args, **kwargs):
 
     scored = []
     for config in configs:
+        block_m = config.kwargs["BLOCK_M"]
         block_n = config.kwargs["BLOCK_N"]
         block_k = config.kwargs["BLOCK_K"]
         # Base: stages * BLOCK_K * (BLOCK_M + BLOCK_N) + BLOCK_M * BLOCK_N
-        smem_base = _estimate_smem_usage(config.num_stages, BLOCK_M, block_n, block_k)
+        smem_base = _estimate_smem_usage(config.num_stages, block_m, block_n, block_k)
         # A tile [BLOCK_R, BLOCK_K] loaded per stage in the inner loop
         smem_lora_loop = config.num_stages * block_r * block_k * 2
         # B tile [BLOCK_N, BLOCK_R] loaded once in epilogue
         smem_lora_epilogue = block_n * block_r * 2
         smem = smem_base + smem_lora_loop + smem_lora_epilogue
+
+        # Register pressure: live tiles are acc[M,N], xa_acc[M,R],
+        # x[M,K], w[K,N], a[R,K], plus epilogue b[N,R]
+        est_regs = _estimate_register_pressure(
+            config.num_warps,
+            (block_m, block_n),  # acc
+            (block_m, block_r),  # xa_acc
+            (block_m, block_k),  # x tile
+            (block_k, block_n),  # w tile
+            (block_r, block_k),  # a tile
+            (block_n, block_r),  # b tile (epilogue)
+        )
+        if est_regs > _MAX_REGS_SOFT_LIMIT:
+            continue
+
         scored.append((smem, config))
 
     pruned = [c for s, c in scored if s <= smem_cap - _SMEM_SLACK]
     if pruned:
         return pruned
-    # All configs exceed SMEM — return the one with smallest estimated usage
-    scored.sort(key=lambda x: x[0])
-    return [scored[0][1]]
+    if scored:
+        # All surviving configs exceed SMEM — return the one with smallest usage
+        scored.sort(key=lambda x: x[0])
+        return [scored[0][1]]
+    # All configs pruned by register pressure — fall back to smallest tiles
+    return [
+        min(
+            configs,
+            key=lambda c: (
+                c.kwargs["BLOCK_M"] * c.kwargs["BLOCK_N"] * c.kwargs["BLOCK_K"]
+            ),
+        )
+    ]
 
 
 @triton.autotune(
@@ -531,6 +587,89 @@ def _scatter2scatter_lora(
     tl.store(Y_blk_ptrs, acc, mask=M_boundary_mask[:, None] & N_mask[None, :])
 
 
+def _scatter2scatter_lora_split(
+    X: torch.Tensor,
+    W: torch.Tensor,
+    sorted_expert_idxs: torch.Tensor,
+    sorted_scattered_idxs: torch.Tensor,
+    k: int,
+    lora_A: torch.Tensor,
+    lora_B: torch.Tensor,
+    scaling: float,
+    b: Optional[torch.Tensor] = None,
+    x_grouped: bool = False,
+    y_grouped: bool = False,
+    out: Optional[torch.Tensor] = None,
+) -> torch.Tensor:
+    """Split base+LoRA forward: 3 scatter2scatter calls, no fused LoRA kernel.
+
+    Faster for models with few large experts (e.g. Mixtral E=8, I=14336)
+    because the base kernel runs at full speed without LoRA SMEM overhead,
+    and the LoRA matmuls (R=16) are tiny separate passes.
+
+    Y = scatter(X, W) + scaling * scatter(scatter(X, A^T), B^T)
+    """
+    from axolotl.integrations.kernels.libs.scattermoe_lora.kernels.ops import (
+        scatter2scatter,
+    )
+
+    E = W.size(0)
+    R = lora_A.size(0) // E
+    K = W.size(1)
+    N = W.size(2)
+
+    # 1. Base: Y_base = X @ W  (uses base kernel with optimal tile sizes)
+    output = scatter2scatter(
+        X=X,
+        W=W,
+        b=b,
+        sorted_expert_idxs=sorted_expert_idxs,
+        sorted_scattered_idxs=sorted_scattered_idxs,
+        k=k,
+        x_grouped=x_grouped,
+        y_grouped=y_grouped,
+        out=out,
+    )
+
+    # 2. XA = X @ A^T  (tiny: output is [M*k, R])
+    # Reshape A: [R*E, K] → [E, K, R] (expert weights for scatter2scatter)
+    W_A = lora_A.reshape(E, R, K).permute(0, 2, 1).contiguous()
+    XA = scatter2scatter(
+        X=X,
+        W=W_A,
+        sorted_expert_idxs=sorted_expert_idxs,
+        sorted_scattered_idxs=sorted_scattered_idxs,
+        k=k,
+        x_grouped=x_grouped,
+        y_grouped=True,
+    )
+
+    # 3. Y_lora = XA @ B^T  (R is tiny, so this is very fast)
+    # Reshape B: [N, R*E] → [E, R, N]
+    W_B = lora_B.T.reshape(E, R, N).contiguous()
+    Y_lora = scatter2scatter(
+        X=XA,
+        W=W_B,
+        sorted_expert_idxs=sorted_expert_idxs,
+        sorted_scattered_idxs=sorted_scattered_idxs,
+        k=1,
+        x_grouped=True,
+        y_grouped=y_grouped,
+    )
+
+    # 4. Y = Y_base + scaling * Y_lora
+    output.add_(Y_lora, alpha=scaling)
+    return output
+
+
+# Threshold for switching from fused to split LoRA forward.
+# Split wins when per-expert matmul is large (bandwidth-bound LoRA tile
+# loads dominate in the fused kernel's inner loop).
+# Empirically: split wins for E<=32 with K*N > 20M (e.g. Mixtral, Phi-MoE).
+_SPLIT_LORA_FWD_THRESHOLD = 20_000_000  # per-expert K*N
+_SPLIT_LORA_FWD_MAX_EXPERTS = 32
+
+
 def scatter2scatter_lora(
     X: torch.Tensor,
     W: torch.Tensor,
@@ -546,7 +685,13 @@ def scatter2scatter_lora(
     out: Optional[torch.Tensor] = None,
 ) -> torch.Tensor:
     """
-    Fused scatter2scatter with LoRA: Y[i] = X[i] @ W[e] + scaling * (X[i] @ A[e]^T) @ B[e]^T + b[e]
+    Scatter2scatter with LoRA: Y[i] = X[i] @ W[e] + scaling * (X[i] @ A[e]^T) @ B[e]^T + b[e]
+
+    Automatically selects between:
+    - Fused kernel: single Triton kernel with LoRA in the inner loop.
+      Best for many small experts (E>=64, small K*N).
+    - Split dispatch: 3 separate scatter2scatter calls (base + XA + lora).
+      Best for few large experts (E<=32, large K*N like Mixtral).
 
     Args:
         X: Input [M, K] or [M*k, K] if x_grouped
@@ -565,12 +710,30 @@ def scatter2scatter_lora(
     Returns:
         Y: Output [M*k, N]
     """
-    assert sorted_scattered_idxs.size(0) == sorted_expert_idxs.size(0)
-    assert sorted_scattered_idxs.size(0) == X.size(0) * k
-
     E = W.size(0)
     K = W.size(1)
     N = W.size(2)
+
+    # Dispatch: split for few large experts, fused for many small experts
+    if E <= _SPLIT_LORA_FWD_MAX_EXPERTS and K * N >= _SPLIT_LORA_FWD_THRESHOLD:
+        return _scatter2scatter_lora_split(
+            X,
+            W,
+            sorted_expert_idxs,
+            sorted_scattered_idxs,
+            k,
+            lora_A,
+            lora_B,
+            scaling,
+            b,
+            x_grouped,
+            y_grouped,
+            out,
+        )
+
+    assert sorted_scattered_idxs.size(0) == sorted_expert_idxs.size(0)
+    assert sorted_scattered_idxs.size(0) == X.size(0) * k
+
     R = lora_A.size(0) // E
 
     # Pad R to power of 2 for Triton tile size
@@ -610,11 +773,9 @@ def scatter2scatter_lora(
         b_ptr,
         stride_be,
         stride_bn,
-        # A: [r*E, K] -> stride(0) is r*E dim stride, stride(1) is K dim stride
         lora_A,
         lora_A.stride(0),
         lora_A.stride(1),
-        # B: [N, r*E] -> stride(0) is N dim stride, stride(1) is r*E dim stride
         lora_B,
         lora_B.stride(0),
         lora_B.stride(1),
@@ -625,9 +786,8 @@ def scatter2scatter_lora(
         K=K,
         N=N,
         E=E,
-        ACTUAL_R=R,  # True LoRA rank for weight indexing
-        BLOCK_M=BLOCK_M,
-        BLOCK_R=BLOCK_R,  # Padded tile size >= max(R, 16)
+        ACTUAL_R=R,
+        BLOCK_R=BLOCK_R,
         ACC_TYPE=tl.float32,
         scaling=scaling,
         allow_tf32=ALLOW_TF32,
@@ -761,13 +921,13 @@ def _compute_expert_block_lora_dX(
         + (A_expert_offset + R_block)[:, None] * stride_ar
         + K_block[None, :] * stride_ak
     )
-    a_e = tl.load(A_blk_ptrs, mask=R_mask[:, None] & K_mask[None, :], other=0.0)
-
-    # Cast to float32 for precision
-    a_f32 = a_e.to(tl.float32)
+    a_e = tl.load(A_blk_ptrs, mask=R_mask[:, None] & K_mask[None, :], other=0.0).to(
+        INPUT_DTYPE
+    )
 
     # (DY @ B) @ A: [M, R] @ [R, K] -> [M, K]
-    lora_dx = tl.dot(dy_b_acc, a_f32, allow_tf32=allow_tf32)
+    # tl.dot requires non-float32 inputs (tensor cores); cast accumulator back to input dtype
+    lora_dx = tl.dot(dy_b_acc.to(INPUT_DTYPE), a_e, allow_tf32=allow_tf32)
 
     acc += scaling * lora_dx
     return acc
@@ -779,25 +939,26 @@ def _scatter2scatter_lora_dX_configs():
     The inner loop is over N (not K as in forward). The output dimension is K.
     So BLOCK_K tiles the output and BLOCK_N tiles the reduction.
 
-    Search space includes smaller tile sizes and fewer pipeline stages to
-    support GPUs with limited shared memory (e.g. ~99KB on some GPUs).
+    BLOCK_M is now autotunable (was fixed at 128).
 
     Search space:
-      BLOCK_K:    {32, 64, 128, 256}   (output tile)
-      BLOCK_N:    {32, 64, 128, 256}   (reduction tile)
+      BLOCK_M:    {32, 64, 128}        (token tile)
+      BLOCK_K:    {32, 64, 128}   (output tile)
+      BLOCK_N:    {32, 64}   (reduction tile)
       num_warps:  {4, 8}
       num_stages: {3, 4, 5}
     """
     configs = []
-    for block_k, block_n, warps, stages in product(
-        [32, 64, 128, 256],  # BLOCK_K (output dimension)
-        [32, 64, 128, 256],  # BLOCK_N (reduction dimension)
+    for block_m, block_k, block_n, warps, stages in product(
+        [32, 64, 128],  # BLOCK_M
+        [32, 64, 128],  # BLOCK_K (output dimension)
+        [32, 64],  # BLOCK_N (reduction dimension)
         [4, 8],  # num_warps
         [3, 4, 5],  # num_stages
     ):
         configs.append(
             triton.Config(
-                {"BLOCK_K": block_k, "BLOCK_N": block_n},
+                {"BLOCK_M": block_m, "BLOCK_K": block_k, "BLOCK_N": block_n},
                 num_stages=stages,
                 num_warps=warps,
             )
@@ -806,7 +967,7 @@ def _scatter2scatter_lora_dX_configs():
 
 
 def _prune_dX_configs(configs, named_args, **kwargs):
-    """Prune backward dX configs based on SMEM capacity.
+    """Prune backward dX configs based on SMEM capacity and register pressure.
 
     The dX kernel inner loop loads three tiles per pipeline stage:
       DY[BLOCK_M, BLOCK_N], W^T[BLOCK_N, BLOCK_K], B[BLOCK_N, BLOCK_R].
@@ -822,23 +983,49 @@ def _prune_dX_configs(configs, named_args, **kwargs):
 
     scored = []
     for config in configs:
+        block_m = config.kwargs["BLOCK_M"]
         block_k = config.kwargs["BLOCK_K"]
         block_n = config.kwargs["BLOCK_N"]
         # Base: stages * BLOCK_N * (BLOCK_M + BLOCK_K) + BLOCK_M * BLOCK_K
-        smem_base = _estimate_smem_usage(config.num_stages, BLOCK_M, block_k, block_n)
+        smem_base = _estimate_smem_usage(config.num_stages, block_m, block_k, block_n)
         # B tile [BLOCK_N, BLOCK_R] loaded per stage in the inner loop
         smem_lora_loop = config.num_stages * block_n * block_r * 2
         # A tile [BLOCK_R, BLOCK_K] loaded once in epilogue
         smem_lora_epilogue = block_r * block_k * 2
         smem = smem_base + smem_lora_loop + smem_lora_epilogue
+
+        # Register pressure: live tiles are acc[M,K], dy_b_acc[M,R],
+        # dy[M,N], wt[N,K], b[N,R], plus epilogue a[R,K]
+        est_regs = _estimate_register_pressure(
+            config.num_warps,
+            (block_m, block_k),  # acc
+            (block_m, block_r),  # dy_b_acc
+            (block_m, block_n),  # dy tile
+            (block_n, block_k),  # wt tile
+            (block_n, block_r),  # b tile
+            (block_r, block_k),  # a tile (epilogue)
+        )
+        if est_regs > _MAX_REGS_SOFT_LIMIT:
+            continue
+
         scored.append((smem, config))
 
     pruned = [c for s, c in scored if s <= smem_cap - _SMEM_SLACK]
     if pruned:
         return pruned
-    # All configs exceed SMEM — return the one with smallest estimated usage
-    scored.sort(key=lambda x: x[0])
-    return [scored[0][1]]
+    if scored:
+        # All surviving configs exceed SMEM — return the one with smallest usage
+        scored.sort(key=lambda x: x[0])
+        return [scored[0][1]]
+    # All configs pruned by register pressure — fall back to smallest tiles
+    return [
+        min(
+            configs,
+            key=lambda c: (
+                c.kwargs["BLOCK_M"] * c.kwargs["BLOCK_K"] * c.kwargs["BLOCK_N"]
+            ),
+        )
+    ]
 
 
 @triton.autotune(
@@ -1067,7 +1254,7 @@ def scatter2scatter_lora_dX(
         N=N,
         E=E,
         ACTUAL_R=R,
-        BLOCK_M=BLOCK_M,
+        # BLOCK_M is autotuned (injected by triton.autotune from Config kwargs)
         BLOCK_R=BLOCK_R,
         ACC_TYPE=tl.float32,
         scaling=scaling,
@@ -1091,9 +1278,9 @@ def _group_bwd_lora_configs():
     support GPUs with limited shared memory (e.g. ~99KB on some GPUs).
 
     Search space:
-      BLOCK_M:    {32, 64, 128, 256}   (token-loop tile)
-      BLOCK_K:    {32, 64, 128, 256}
-      BLOCK_N:    {32, 64, 128, 256}
+      BLOCK_M:    {32, 64, 128}   (token-loop tile)
+      BLOCK_K:    {32, 64, 128}
+      BLOCK_N:    {32, 64}
       num_warps:  {4, 8}
       num_stages: {3, 4, 5}
 
@@ -1102,9 +1289,9 @@ def _group_bwd_lora_configs():
     """
     configs = []
     for block_m, block_k, block_n, warps, stages in product(
-        [32, 64, 128, 256],  # BLOCK_M
-        [32, 64, 128, 256],  # BLOCK_K
-        [32, 64, 128, 256],  # BLOCK_N
+        [32, 64, 128],  # BLOCK_M
+        [32, 64, 128],  # BLOCK_K
+        [32, 64],  # BLOCK_N
         [4, 8],  # num_warps
         [3, 4, 5],  # num_stages
     ):
@@ -1119,7 +1306,7 @@ def _group_bwd_lora_configs():
 
 
 def _prune_bwd_lora_configs(configs, named_args, **kwargs):
-    """Prune backward configs based on SMEM capacity.
+    """Prune backward configs based on SMEM capacity and register pressure.
 
     The backward kernel loads X[BLOCK_M, BLOCK_K] and DY[BLOCK_M, BLOCK_N]
     in the inner loop, plus holds A[BLOCK_R, BLOCK_K] and B[BLOCK_N, BLOCK_R]
@@ -1138,14 +1325,40 @@ def _prune_bwd_lora_configs(configs, named_args, **kwargs):
         # A[BLOCK_R, BLOCK_K] and B[BLOCK_N, BLOCK_R] held for the full expert
         smem_lora = (block_r * block_k + block_n * block_r) * 2
         smem = smem_base + smem_lora
+
+        # Register pressure: dA_acc[R,K], dB_acc[N,R], x[M,K], dy[M,N],
+        # a[R,K], b[N,R], xa[M,R], dy_b[M,R]
+        est_regs = _estimate_register_pressure(
+            config.num_warps,
+            (block_r, block_k),  # dA_acc
+            (block_n, block_r),  # dB_acc
+            (block_m, block_k),  # x tile
+            (block_m, block_n),  # dy tile
+            (block_r, block_k),  # a tile
+            (block_n, block_r),  # b tile
+            (block_m, block_r),  # xa intermediate
+        )
+        if est_regs > _MAX_REGS_SOFT_LIMIT:
+            continue
+
         scored.append((smem, config))
 
     pruned = [c for s, c in scored if s <= smem_cap - _SMEM_SLACK]
     if pruned:
         return pruned
-    # All configs exceed SMEM — return the one with smallest estimated usage
-    scored.sort(key=lambda x: x[0])
-    return [scored[0][1]]
+    if scored:
+        # All surviving configs exceed SMEM — return the one with smallest usage
+        scored.sort(key=lambda x: x[0])
+        return [scored[0][1]]
+    # All configs pruned by register pressure — fall back to smallest tiles
+    return [
+        min(
+            configs,
+            key=lambda c: (
+                c.kwargs["BLOCK_M"] * c.kwargs["BLOCK_K"] * c.kwargs["BLOCK_N"]
+            ),
+        )
+    ]
 
 
 @triton.autotune(
@@ -1330,6 +1543,279 @@ def _group_bwd_lora(
         )
 
 
+def _group_bwd_split_configs():
+    """Autotune configs for split dA/dB kernels."""
+    configs = []
+    for block_m, block_dim, warps, stages in product(
+        [32, 64, 128],  # BLOCK_M (token tile)
+        [32, 64, 128, 256],  # BLOCK_DIM (K for dA, N for dB — output tile)
+        [4, 8],  # num_warps
+        [3, 4, 5],  # num_stages
+    ):
+        configs.append(
+            triton.Config(
+                {"BLOCK_M": block_m, "BLOCK_DIM": block_dim},
+                num_stages=stages,
+                num_warps=warps,
+            )
+        )
+    return configs
+
+
+def _prune_split_configs(configs, named_args, **kwargs):
+    """Prune split kernel configs based on SMEM capacity and register pressure."""
+    smem_cap = _get_smem_capacity()
+    block_r = named_args.get("BLOCK_R", 64)
+
+    # Fixed inner tile for reduction dimension
+    BLOCK_INNER = 64
+
+    pruned = []
+    for config in configs:
+        block_m = config.kwargs["BLOCK_M"]
+        block_dim = config.kwargs["BLOCK_DIM"]
+        # Inner loop loads: input[M, INNER] and other[M, INNER_or_DIM]
+        smem = config.num_stages * BLOCK_INNER * (block_m + block_dim) * 2
+        # LoRA weights held in registers: [INNER, R] or [R, DIM]
+        smem += (block_r * max(block_dim, BLOCK_INNER)) * 2
+
+        # Register pressure check
+        est_regs = _estimate_register_pressure(
+            config.num_warps,
+            (block_r, block_dim),  # acc
+            (block_m, BLOCK_INNER),  # input tile
+            (block_m, block_dim),  # other tile
+            (block_r, BLOCK_INNER),  # lora weight
+        )
+        if est_regs > _MAX_REGS_SOFT_LIMIT:
+            continue
+
+        if smem <= smem_cap - _SMEM_SLACK:
+            pruned.append(config)
+
+    if pruned:
+        return pruned
+    configs.sort(key=lambda c: c.kwargs["BLOCK_M"] * c.kwargs["BLOCK_DIM"])
+    return [configs[0]]
+
+
+@triton.autotune(
+    configs=_group_bwd_split_configs(),
+    key=["M", "K", "N"],
+    prune_configs_by={"early_config_prune": _prune_split_configs},
+)
+@triton.heuristics(
+    {
+        "NO_DIM_MASK": lambda args: (
+            (args["K"] % args["BLOCK_DIM"]) == 0
+            if args["COMPUTE_DA"]
+            else (args["N"] % args["BLOCK_DIM"]) == 0
+        ),
+    }
+)
+@triton.jit
+def _group_bwd_lora_split(
+    # Data tensors (DY and X are always present)
+    DY_ptr,
+    stride_dym,
+    stride_dyn,
+    X_ptr,
+    stride_xm,
+    stride_xk,
+    # LoRA weight for the inner reduction (B for dA, A for dB)
+    LW_ptr,
+    stride_lw0,
+    stride_lw1,
+    # Output gradient tensor (dA or dB)
+    OUT_ptr,
+    stride_out0,
+    stride_out1,
+    # Expert offsets
+    expert_offsets_ptr,
+    # Dimensions
+    M,
+    K: tl.constexpr,
+    N: tl.constexpr,
+    ACTUAL_R: tl.constexpr,
+    BLOCK_R: tl.constexpr,
+    INNER_DIM: tl.constexpr,  # reduction dimension (N for dA, K for dB)
+    scaling,
+    # Mode flag
+    COMPUTE_DA: tl.constexpr,  # True = compute dA, False = compute dB
+    # Tile sizes
+    BLOCK_M: tl.constexpr,
+    BLOCK_DIM: tl.constexpr,
+    ACC_TYPE: tl.constexpr,
+    allow_tf32: tl.constexpr,
+    NO_DIM_MASK: tl.constexpr,
+):
+    """
+    Unified split kernel for LoRA gradient computation.
+
+    When COMPUTE_DA=True:
+      dA[e] = scaling * (dY @ B[e])^T @ X  →  [R, K]
+      Grid: (E, cdiv(K, BLOCK_DIM))
+      - outer_ptr/stride = X (read [M, K_block])
+      - inner reduction over N using DY and B
+      - output shape [BLOCK_R, BLOCK_DIM]
+
+    When COMPUTE_DA=False:
+      dB[e] = scaling * dY^T @ (X @ A[e]^T)  →  [N, R]
+      Grid: (E, cdiv(N, BLOCK_DIM))
+      - outer_ptr/stride = DY (read [M, N_block])
+      - inner reduction over K using X and A
+      - output shape [BLOCK_DIM, BLOCK_R]
+
+    No atomic adds — each (E, dim_block) pair is written by exactly one block.
+    """
+    E_idx = tl.program_id(0)
+    dim_block_id = tl.program_id(1)
+
+    if E_idx == 0:
+        start_idx = 0
+    else:
+        start_idx = tl.load(expert_offsets_ptr + E_idx - 1).to(tl.int32)
+    end_idx = tl.load(expert_offsets_ptr + E_idx).to(tl.int32)
+    num_tokens = end_idx - start_idx
+
+    # Output dimension tile (K for dA, N for dB)
+    if COMPUTE_DA:
+        OUT_DIM: tl.constexpr = K  # type: ignore[no-redef]
+    else:
+        OUT_DIM: tl.constexpr = N  # type: ignore[no-redef]
+    dim_block = dim_block_id * BLOCK_DIM + tl.arange(0, BLOCK_DIM)
+    dim_mask = dim_block < OUT_DIM
+    R_block = tl.arange(0, BLOCK_R)
+    R_mask = R_block < ACTUAL_R
+    lora_offset = E_idx * ACTUAL_R
+
+    # Output pointers — layout differs: dA is [R, K], dB is [N, R]
+    if COMPUTE_DA:
+        out_blk_ptrs = (
+            OUT_ptr
+            + (lora_offset + R_block)[:, None] * stride_out0
+            + dim_block[None, :] * stride_out1
+        )
+        out_mask = R_mask[:, None] & dim_mask[None, :]
+    else:
+        out_blk_ptrs = (
+            OUT_ptr
+            + dim_block[:, None] * stride_out0
+            + (lora_offset + R_block)[None, :] * stride_out1
+        )
+        out_mask = dim_mask[:, None] & R_mask[None, :]
+
+    if num_tokens > 0:
+        M_block = tl.arange(0, BLOCK_M)
+        INPUT_DTYPE = X_ptr.dtype.element_ty
+        BLOCK_INNER: tl.constexpr = 64
+        inner_iters = tl.cdiv(INNER_DIM, BLOCK_INNER)
+
+        if COMPUTE_DA:
+            acc = tl.zeros((BLOCK_R, BLOCK_DIM), dtype=ACC_TYPE)
+        else:
+            acc = tl.zeros((BLOCK_DIM, BLOCK_R), dtype=ACC_TYPE)
+
+        M_iters = tl.cdiv(num_tokens, BLOCK_M)
+        for i in range(M_iters):
+            M_idx = start_idx + i * BLOCK_M + M_block
+            M_mask = M_idx < end_idx
+
+            if COMPUTE_DA:
+                # Load X[M, K_block] (the "outer" tensor for dA)
+                outer = tl.load(
+                    X_ptr + M_idx[:, None] * stride_xm + dim_block[None, :] * stride_xk,
+                    mask=M_mask[:, None] & dim_mask[None, :],
+                    other=0.0,
+                ).to(INPUT_DTYPE)
+
+                # Reduce DY[M, :] @ B[e][:, R] over N → [M, R]
+                reduced = tl.zeros((BLOCK_M, BLOCK_R), dtype=ACC_TYPE)
+                inner_range = tl.arange(0, BLOCK_INNER)
+                for j in range(inner_iters):
+                    inn_off = j * BLOCK_INNER + inner_range
+                    inn_mask = inn_off < N
+
+                    dy_tile = tl.load(
+                        DY_ptr
+                        + M_idx[:, None] * stride_dym
+                        + inn_off[None, :] * stride_dyn,
+                        mask=M_mask[:, None] & inn_mask[None, :],
+                        other=0.0,
+                    ).to(INPUT_DTYPE)
+                    # B layout: [N, r*E] → stride_lw0=N stride, stride_lw1=r*E stride
+                    lw_tile = tl.load(
+                        LW_ptr
+                        + inn_off[:, None] * stride_lw0
+                        + (lora_offset + R_block)[None, :] * stride_lw1,
+                        mask=inn_mask[:, None] & R_mask[None, :],
+                        other=0.0,
+                    ).to(INPUT_DTYPE)
+                    reduced += tl.dot(dy_tile, lw_tile, allow_tf32=allow_tf32)
+
+                # dA += (DY@B)^T @ X: [R, M] @ [M, K_block] → [R, K_block]
+                acc += tl.dot(
+                    tl.trans(reduced.to(INPUT_DTYPE)), outer, allow_tf32=allow_tf32
+                )
+            else:
+                # Load DY[M, N_block] (the "outer" tensor for dB)
+                outer = tl.load(
+                    DY_ptr
+                    + M_idx[:, None] * stride_dym
+                    + dim_block[None, :] * stride_dyn,
+                    mask=M_mask[:, None] & dim_mask[None, :],
+                    other=0.0,
+                ).to(INPUT_DTYPE)
+
+                # Reduce X[M, :] @ A[e][:, :].T over K → [M, R]
+                reduced = tl.zeros((BLOCK_M, BLOCK_R), dtype=ACC_TYPE)
+                inner_range = tl.arange(0, BLOCK_INNER)
+                for j in range(inner_iters):
+                    inn_off = j * BLOCK_INNER + inner_range
+                    inn_mask = inn_off < K
+
+                    x_tile = tl.load(
+                        X_ptr
+                        + M_idx[:, None] * stride_xm
+                        + inn_off[None, :] * stride_xk,
+                        mask=M_mask[:, None] & inn_mask[None, :],
+                        other=0.0,
+                    ).to(INPUT_DTYPE)
+                    # A layout: [r*E, K] → stride_lw0=r*E stride, stride_lw1=K stride
+                    # We want A[e]^T: [K, R], so load as [K_inner, R]
+                    lw_tile = tl.load(
+                        LW_ptr
+                        + (lora_offset + R_block)[None, :] * stride_lw0
+                        + inn_off[:, None] * stride_lw1,
+                        mask=inn_mask[:, None] & R_mask[None, :],
+                        other=0.0,
+                    ).to(INPUT_DTYPE)
+                    reduced += tl.dot(x_tile, lw_tile, allow_tf32=allow_tf32)
+
+                # dB += DY^T @ (X@A^T): [N_block, M] @ [M, R] → [N_block, R]
+                acc += tl.dot(
+                    tl.trans(outer), reduced.to(INPUT_DTYPE), allow_tf32=allow_tf32
+                )
+
+        tl.store(
+            out_blk_ptrs, (acc * scaling).to(OUT_ptr.dtype.element_ty), mask=out_mask
+        )
+    else:
+        # Zero out this expert's slice — needed because output uses empty_like
+        if COMPUTE_DA:
+            tl.store(
+                out_blk_ptrs,
+                tl.zeros((BLOCK_R, BLOCK_DIM), dtype=OUT_ptr.dtype.element_ty),
+                mask=out_mask,
+            )
+        else:
+            tl.store(
+                out_blk_ptrs,
+                tl.zeros((BLOCK_DIM, BLOCK_R), dtype=OUT_ptr.dtype.element_ty),
+                mask=out_mask,
+            )
+
+
 def group_bwd_lora(
     DY: torch.Tensor,
     X: torch.Tensor,
@@ -1344,6 +1830,9 @@ def group_bwd_lora(
     """
     Compute LoRA gradients for A and B on expert-grouped data.
 
+    Uses split dA/dB kernels that eliminate atomic adds by giving each
+    (expert, output_block) pair its own thread block.
+
     Args:
         DY: Gradient w.r.t. output [M_total, N] (grouped by expert)
         X: Input [M_total, K] (grouped by expert)
@@ -1361,19 +1850,46 @@ def group_bwd_lora(
     K = X.size(1)
     N = DY.size(1)
 
-    # Zero-init for atomic accumulation
-    dA = torch.zeros_like(lora_A)
-    dB = torch.zeros_like(lora_B)
+    # No zero-init needed: the split kernels write zeros for experts with
+    # zero routed tokens directly in the kernel (else branch).
+    dA = torch.empty_like(lora_A)
+    dB = torch.empty_like(lora_B)
 
     BLOCK_R = _block_r_for_rank(R)
 
-    def grid(META):
-        return (
-            E * triton.cdiv(K, META["BLOCK_K"]),
-            triton.cdiv(N, META["BLOCK_N"]),
-        )
+    def grid_dA(META):
+        return (E, triton.cdiv(K, META["BLOCK_DIM"]))
 
-    _group_bwd_lora[grid](
+    _group_bwd_lora_split[grid_dA](
+        DY,
+        DY.stride(0),
+        DY.stride(1),
+        X,
+        X.stride(0),
+        X.stride(1),
+        lora_B,
+        lora_B.stride(0),
+        lora_B.stride(1),
+        dA,
+        dA.stride(0),
+        dA.stride(1),
+        expert_offsets,
+        M=DY.size(0),
+        K=K,
+        N=N,
+        ACTUAL_R=R,
+        BLOCK_R=BLOCK_R,
+        INNER_DIM=N,
+        scaling=scaling,
+        COMPUTE_DA=True,
+        ACC_TYPE=tl.float32,
+        allow_tf32=ALLOW_TF32,
+    )
+
+    def grid_dB(META):
+        return (E, triton.cdiv(N, META["BLOCK_DIM"]))
+
+    _group_bwd_lora_split[grid_dB](
         DY,
         DY.stride(0),
         DY.stride(1),
@@ -1383,12 +1899,6 @@ def group_bwd_lora(
         lora_A,
         lora_A.stride(0),
         lora_A.stride(1),
-        lora_B,
-        lora_B.stride(0),
-        lora_B.stride(1),
-        dA,
-        dA.stride(0),
-        dA.stride(1),
         dB,
         dB.stride(0),
         dB.stride(1),
@@ -1396,9 +1906,11 @@ def group_bwd_lora(
         M=DY.size(0),
         K=K,
         N=N,
-        ACTUAL_R=R,  # True LoRA rank
-        BLOCK_R=BLOCK_R,  # Padded tile size
+        ACTUAL_R=R,
+        BLOCK_R=BLOCK_R,
+        INNER_DIM=K,
         scaling=scaling,
+        COMPUTE_DA=False,
         ACC_TYPE=tl.float32,
         allow_tf32=ALLOW_TF32,
     )

src/axolotl/integrations/kernels/libs/scattermoe_lora/layers.py

@@ -489,20 +489,71 @@ class HFScatterMoEGatedMLP(nn.Module):
         # ====================================================================
         experts, gup_lora, down_lora = _unwrap_experts_lora(self.experts)
 
+        # ====================================================================
+        # Selective expert weight dequantization
+        # ====================================================================
+        # When experts are BnB-quantized (quantize_moe_experts), dequantize
+        # only the active experts instead of all E. This saves ~97% memory
+        # for the transient dequant buffer when few experts are active.
+        use_selective = (
+            getattr(self, "_use_selective_dequant", False)
+            and hasattr(experts, "parametrizations")
+            and "gate_up_proj" in experts.parametrizations
+        )
+
+        if use_selective:
+            from axolotl.integrations.kernels.libs.scattermoe_lora.selective_dequant import (
+                get_active_experts,
+                remap_expert_indices,
+                selective_expert_weights,
+                selective_lora_weights,
+            )
+
+            active_experts = get_active_experts(sorted_expert_idxs, num_experts)
+            remapped_expert_idxs, compact_offsets = remap_expert_indices(
+                sorted_expert_idxs,
+                expert_offsets,
+                active_experts,
+                num_experts,
+            )
+            # Dequantize only active experts' weights
+            gate_up_W = selective_expert_weights(
+                experts,
+                "gate_up_proj",
+                active_experts,
+            ).transpose(2, 1)  # [num_active, hidden, 2*inter]
+
+            # Remap LoRA weights to match compact expert indices
+            if gup_lora is not None:
+                gup_A, gup_B, gup_scaling = gup_lora
+                gup_A, gup_B = selective_lora_weights(
+                    gup_A,
+                    gup_B,
+                    active_experts,
+                    num_experts,
+                )
+                gup_lora = (gup_A, gup_B, gup_scaling)
+
+            # Use remapped indices for ScatterMoE kernels
+            sei_gup = remapped_expert_idxs
+            eo_gup = compact_offsets
+        else:
+            gate_up_W = experts.gate_up_proj.transpose(2, 1)  # [E, hidden, 2*inter]
+            sei_gup = sorted_expert_idxs
+            eo_gup = expert_offsets
+
         # ====================================================================
         # Gate + Up projection
         # ====================================================================
-        gate_up_W = experts.gate_up_proj.transpose(2, 1)  # [E, hidden, 2*inter]
-
         if gup_lora is not None:
             gup_A, gup_B, gup_scaling = gup_lora
             gup = parallel_linear_lora(
                 hidden_states_flat,
                 gate_up_W,
                 top_k,
-                sorted_expert_idxs,
+                sei_gup,
                 sorted_scattered_idxs,
-                expert_offsets,
+                eo_gup,
                 lora_A=gup_A,
                 lora_B=gup_B,
                 scaling=gup_scaling,
@@ -516,9 +567,9 @@ class HFScatterMoEGatedMLP(nn.Module):
                 hidden_states_flat,
                 gate_up_W,
                 top_k,
-                sorted_expert_idxs,
+                sei_gup,
                 sorted_scattered_idxs,
-                expert_offsets,
+                eo_gup,
                 grouped_in=False,
                 grouped_out=True,
             )
@@ -529,7 +580,29 @@ class HFScatterMoEGatedMLP(nn.Module):
         # ====================================================================
         # Down projection
         # ====================================================================
-        down_W = experts.down_proj.transpose(2, 1)  # [E, inter, hidden]
+        if use_selective:
+            down_W = selective_expert_weights(
+                experts,
+                "down_proj",
+                active_experts,
+            ).transpose(2, 1)  # [num_active, inter, hidden]
+
+            if down_lora is not None:
+                down_A, down_B, down_scaling = down_lora
+                down_A, down_B = selective_lora_weights(
+                    down_A,
+                    down_B,
+                    active_experts,
+                    num_experts,
+                )
+                down_lora = (down_A, down_B, down_scaling)
+
+            sei_down = remapped_expert_idxs
+            eo_down = compact_offsets
+        else:
+            down_W = experts.down_proj.transpose(2, 1)  # [E, inter, hidden]
+            sei_down = sorted_expert_idxs
+            eo_down = expert_offsets
 
         if down_lora is not None:
             down_A, down_B, down_scaling = down_lora
@@ -537,9 +610,9 @@ class HFScatterMoEGatedMLP(nn.Module):
                 h,
                 down_W,
                 1,
-                sorted_expert_idxs,
+                sei_down,
                 sorted_scattered_idxs,
-                expert_offsets,
+                eo_down,
                 lora_A=down_A,
                 lora_B=down_B,
                 scaling=down_scaling,
@@ -554,9 +627,9 @@ class HFScatterMoEGatedMLP(nn.Module):
                 h,
                 down_W,
                 1,
-                sorted_expert_idxs,
+                sei_down,
                 sorted_scattered_idxs,
-                expert_offsets,
+                eo_down,
                 grouped_in=True,
                 grouped_out=False,
                 gates=routing_weights,

src/axolotl/integrations/kernels/libs/scattermoe_lora/selective_dequant.py

@@ -0,0 +1,282 @@
+"""
+Selective Expert Dequantization
+===============================
+
+Instead of dequantizing all E expert weight matrices at once (which creates
+a ~1 GB transient buffer for 256 experts), only dequantize the experts that
+are actually routed to by the current batch's top-k selection.
+
+For Qwen3.5-35B-A3B (E=256, top_k=8, hidden=2048, intermediate=512):
+  - Full dequant: [256, 2048, 1024] = 1,074 MB per projection
+  - Selective (8 active): [8, 2048, 1024] = 33.5 MB per projection
+  - Savings: ~97% memory reduction per layer
+
+This module provides format-agnostic selective weight extraction:
+  - BnB 4-bit (nf4/fp4): slice quantized data + absmax per expert
+  - bf16/fp32: direct indexing (no dequant needed)
+  - FP8: slice + cast
+
+The ScatterMoE kernel itself doesn't change — we remap expert indices
+from global (0..E-1) to compact (0..num_active-1) and pass the smaller
+weight tensor.
+"""
+
+import torch
+import torch.nn as nn
+
+
+def get_active_experts(sorted_expert_idxs: torch.Tensor, E: int) -> torch.Tensor:
+    """Get sorted unique expert indices from the routing output.
+
+    Args:
+        sorted_expert_idxs: Expert assignments sorted by expert id [T*k]
+        E: Total number of experts
+
+    Returns:
+        active: Sorted unique expert indices [num_active]
+    """
+    return torch.unique(sorted_expert_idxs)
+
+
+def remap_expert_indices(
+    sorted_expert_idxs: torch.Tensor,
+    expert_offsets: torch.Tensor,
+    active_experts: torch.Tensor,
+    E: int,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Remap global expert indices to compact indices.
+
+    Maps expert ids from [0..E-1] to [0..num_active-1], preserving the
+    sort order. Also compacts expert_offsets to only active experts.
+
+    Args:
+        sorted_expert_idxs: [T*k] expert ids in sorted order
+        expert_offsets: [E] cumulative token counts (original)
+        active_experts: [num_active] sorted unique expert ids
+        E: Total number of experts
+
+    Returns:
+        remapped_idxs: [T*k] expert ids in [0..num_active-1]
+        compact_offsets: [num_active] cumulative token counts
+    """
+    # Build remap table: global_id -> compact_id
+    remap = torch.empty(E, dtype=torch.long, device=sorted_expert_idxs.device)
+    remap[active_experts] = torch.arange(
+        len(active_experts), device=sorted_expert_idxs.device
+    )
+
+    remapped_idxs = remap[sorted_expert_idxs]
+
+    # Compact the expert_offsets: only keep active experts' cumulative counts
+    compact_offsets = expert_offsets[active_experts]
+
+    return remapped_idxs, compact_offsets
+
+
+def _selective_dequant_bnb4(
+    raw_param: torch.Tensor,
+    quant_state,
+    active_experts: torch.Tensor,
+    expert_shape: tuple[int, int],
+) -> torch.Tensor:
+    """Dequantize only selected experts from BnB 4-bit packed data.
+
+    The raw parameter is a flattened 4-bit packed tensor. Each expert's
+    data is contiguous (stored in expert-major order), so we can gather
+    the packed data and absmax blocks for active experts, then dequantize
+    as one contiguous block.
+
+    Args:
+        raw_param: Flattened uint8 tensor of packed 4-bit weights
+        quant_state: BnB QuantState with absmax, blocksize, code, etc.
+        active_experts: [num_active] expert indices to dequantize
+        expert_shape: (dim1, dim2) shape per expert (e.g. (1024, 2048))
+
+    Returns:
+        Dequantized weights [num_active, dim1, dim2] in original dtype
+    """
+    import bitsandbytes.functional as F  # noqa: N812
+    from bitsandbytes.functional import QuantState
+
+    expert_numel = expert_shape[0] * expert_shape[1]
+    packed_per_expert = expert_numel // 2  # 4-bit = 2 values per byte
+    blocks_per_expert = expert_numel // quant_state.blocksize
+    num_active = len(active_experts)
+
+    if blocks_per_expert == 0:
+        # Expert is smaller than one quantization block — blocks span across
+        # expert boundaries, so per-expert slicing isn't possible.
+        # Fallback: full dequantize + index.
+        full = F.dequantize_4bit(raw_param, quant_state)
+        E_total = full.numel() // expert_numel
+        return full.reshape(E_total, *expert_shape)[active_experts]
+
+    # Use fused Triton kernel for NF4 (handles selective gather + dequant in one pass)
+    if quant_state.quant_type == "nf4" and raw_param.dtype == torch.uint8:
+        from axolotl.integrations.kernels.libs.scattermoe_lora.selective_dequant_kernel import (
+            selective_dequant_nf4_triton,
+        )
+
+        # Handle nested (double) quantization: dequantize absmax first
+        # BnB uses dequantize_blockwise (not _4bit) for nested absmax + offset
+        if quant_state.nested:
+            absmax = F.dequantize_blockwise(quant_state.absmax, quant_state.state2)
+            absmax += quant_state.offset
+            if absmax.dtype != torch.float32:
+                absmax = absmax.float()
+        else:
+            absmax = quant_state.absmax
+
+        return selective_dequant_nf4_triton(
+            packed_data=raw_param,
+            absmax=absmax,
+            active_experts=active_experts,
+            expert_shape=expert_shape,
+            blocksize=quant_state.blocksize,
+            dtype=quant_state.dtype,
+            codebook=quant_state.code,
+        )
+
+    # Fallback: gather + BnB dequant (for fp4 or non-uint8 packed formats)
+    raw_flat = raw_param.reshape(-1)
+
+    offsets_qt = (
+        active_experts.long()[:, None] * packed_per_expert
+        + torch.arange(packed_per_expert, device=raw_param.device)[None, :]
+    ).reshape(-1)
+    qt_gathered = raw_flat[offsets_qt]
+
+    offsets_abs = (
+        active_experts.long()[:, None] * blocks_per_expert
+        + torch.arange(blocks_per_expert, device=raw_param.device)[None, :]
+    ).reshape(-1)
+
+    if quant_state.nested:
+        full_absmax = F.dequantize_blockwise(quant_state.absmax, quant_state.state2)
+        full_absmax += quant_state.offset
+        if full_absmax.dtype != torch.float32:
+            full_absmax = full_absmax.float()
+        absmax_gathered = full_absmax[offsets_abs]
+    else:
+        absmax_gathered = quant_state.absmax[offsets_abs]
+
+    qt_gathered = qt_gathered.unsqueeze(1) if qt_gathered.dim() == 1 else qt_gathered
+
+    gathered_qs = QuantState(
+        absmax=absmax_gathered,
+        shape=torch.Size([num_active * expert_numel]),
+        blocksize=quant_state.blocksize,
+        quant_type=quant_state.quant_type,
+        code=quant_state.code,
+        dtype=quant_state.dtype,
+    )
+
+    deq = F.dequantize_4bit(qt_gathered, gathered_qs)
+    return deq.reshape(num_active, *expert_shape)
+
+
+def _selective_index_dense(
+    param: torch.Tensor,
+    active_experts: torch.Tensor,
+) -> torch.Tensor:
+    """Select experts from a dense (bf16/fp32) weight tensor.
+
+    Simple indexing — no dequantization needed.
+    """
+    return param[active_experts]
+
+
+def selective_expert_weights(
+    experts_module: nn.Module,
+    param_name: str,
+    active_experts: torch.Tensor,
+) -> torch.Tensor:
+    """Extract and dequantize only the active experts' weights.
+
+    Format-agnostic: dispatches based on whether the parameter is
+    BnB 4-bit quantized (via parametrize), FP8, or dense bf16/fp32.
+
+    Args:
+        experts_module: The base experts module (e.g. Qwen3_5MoeExperts)
+        param_name: "gate_up_proj" or "down_proj"
+        active_experts: [num_active] sorted unique expert indices
+
+    Returns:
+        Compact weight tensor [num_active, dim1, dim2] ready for ScatterMoE
+    """
+    # Check if the parameter is BnB-quantized via parametrize
+    if (
+        hasattr(experts_module, "parametrizations")
+        and param_name in experts_module.parametrizations
+    ):
+        param_list = experts_module.parametrizations[param_name]
+        parametrization = param_list[0]
+
+        # BnB 4-bit parametrization
+        if hasattr(parametrization, "quant_state"):
+            # The raw quantized data is on the ParametrizationList, not the
+            # individual Bnb4bitParametrization module
+            raw_param = param_list.original
+            qs = parametrization.quant_state
+            # qs.shape is the original tensor shape before flattening.
+            # For MoE experts it's [E, d1, d2] (3D) or [total_elements] (1D).
+            orig_shape = qs.shape
+            if isinstance(orig_shape, torch.Size) and len(orig_shape) == 3:
+                expert_shape = (orig_shape[1], orig_shape[2])
+            elif isinstance(orig_shape, torch.Size) and len(orig_shape) == 1:
+                # Flattened — need to infer from module attributes
+                E_total = getattr(experts_module, "num_experts", None)
+                if E_total is None:
+                    E_total = int(active_experts.max().item()) + 1
+                expert_numel = orig_shape[0] // E_total
+                d2 = getattr(experts_module, "hidden_dim", None) or getattr(
+                    experts_module, "intermediate_dim", None
+                )
+                if d2 and expert_numel % d2 == 0:
+                    expert_shape = (expert_numel // d2, d2)
+                else:
+                    full = getattr(experts_module, param_name)
+                    return full[active_experts]
+            else:
+                full = getattr(experts_module, param_name)
+                return full[active_experts]
+
+            return _selective_dequant_bnb4(raw_param, qs, active_experts, expert_shape)
+
+    # Dense parameter (bf16/fp32) — direct indexing
+    param = getattr(experts_module, param_name)
+    if param.dim() == 3:
+        return param[active_experts]
+
+    # Fallback: full access
+    return param
+
+
+def selective_lora_weights(
+    lora_A: torch.Tensor,
+    lora_B: torch.Tensor,
+    active_experts: torch.Tensor,
+    E: int,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Select LoRA A and B weights for only the active experts.
+
+    LoRA layout (scattermoe format):
+      A: [r*E, K] — expert e occupies rows [e*r : (e+1)*r]
+      B: [N, r*E] — expert e occupies cols [e*r : (e+1)*r]
+
+    Returns compact:
+      A: [r*num_active, K]
+      B: [N, r*num_active]
+    """
+    R = lora_A.size(0) // E
+
+    # Vectorized gather: active_experts[:, None] * R + arange(R)[None, :]
+    row_idx = (
+        active_experts.long()[:, None] * R
+        + torch.arange(R, device=lora_A.device)[None, :]
+    ).reshape(-1)
+
+    compact_A = lora_A[row_idx]  # [r*num_active, K]
+    compact_B = lora_B[:, row_idx]  # [N, r*num_active]
+
+    return compact_A, compact_B

src/axolotl/integrations/kernels/libs/scattermoe_lora/selective_dequant_kernel.py

@@ -0,0 +1,179 @@
+"""
+Triton kernel for fused selective expert gather + NF4 dequantization.
+
+Instead of:
+  1. Gather packed uint8 data for active experts (memory copy)
+  2. Gather absmax for active experts (memory copy)
+  3. Call BnB dequantize_4bit CUDA kernel
+
+This kernel does all three in one pass:
+  - Reads packed NF4 bytes from expert-strided positions
+  - Looks up the NF4 codebook
+  - Multiplies by the per-block absmax
+  - Writes bf16 output directly
+
+This eliminates the intermediate gather buffer entirely.
+"""
+
+import torch
+import triton
+import triton.language as tl
+
+# NF4 codebook (16 values, precomputed by BnB)
+# These are the normalized float4 reconstruction values
+NF4_CODEBOOK = [
+    -1.0,
+    -0.6961928009986877,
+    -0.5250730514526367,
+    -0.39491748809814453,
+    -0.28444138169288635,
+    -0.18477343022823334,
+    -0.09105003625154495,
+    0.0,
+    0.07958029955625534,
+    0.16093020141124725,
+    0.24611230194568634,
+    0.33791524171829224,
+    0.44070982933044434,
+    0.5626170039176941,
+    0.7229568362236023,
+    1.0,
+]
+
+
+@triton.jit
+def _selective_dequant_nf4_kernel(
+    # Input: packed NF4 data (flattened, expert-major order)
+    packed_ptr,
+    # Input: absmax values (flattened, expert-major order)
+    absmax_ptr,
+    # Input: active expert indices
+    active_experts_ptr,
+    # Input: NF4 codebook (16 float values)
+    codebook_ptr,
+    # Output: dequantized bf16 weights [num_active, expert_numel]
+    out_ptr,
+    stride_out_e,  # stride for expert dim in output
+    # Dimensions
+    num_active,
+    packed_per_expert,  # expert_numel // 2
+    blocks_per_expert,  # expert_numel // blocksize
+    blocksize: tl.constexpr,
+    # Tile size
+    BLOCK_SIZE: tl.constexpr,  # elements per thread block (must be multiple of 2)
+):
+    """
+    Each program processes BLOCK_SIZE elements from one expert.
+
+    Grid: (num_active, cdiv(expert_numel, BLOCK_SIZE))
+
+    For each output element:
+      1. Compute which byte in packed data contains this element
+      2. Extract the 4-bit nibble (high or low)
+      3. Look up in NF4 codebook
+      4. Scale by absmax for this block
+    """
+    expert_local_idx = tl.program_id(0)  # which active expert (0..num_active-1)
+    block_id = tl.program_id(1)  # which element block
+
+    # Load the global expert index
+    expert_global = tl.load(active_experts_ptr + expert_local_idx).to(tl.int64)
+
+    expert_numel = packed_per_expert * 2  # 2 elements per packed byte
+    elem_offset = block_id * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
+    mask = elem_offset < expert_numel
+
+    # Each element is packed as: byte[i//2], low nibble for even i, high for odd i
+    byte_idx = elem_offset // 2
+    is_high = (elem_offset % 2) == 1
+
+    # Read packed bytes from the global expert's region
+    packed_global_offset = expert_global * packed_per_expert + byte_idx
+    packed_bytes = tl.load(packed_ptr + packed_global_offset, mask=mask, other=0).to(
+        tl.int32
+    )
+
+    # Extract 4-bit nibble
+    # BnB packing: high nibble = even element, low nibble = odd element
+    nibble = tl.where(is_high, packed_bytes & 0xF, (packed_bytes >> 4) & 0xF)
+
+    # NF4 codebook lookup
+    # Load all 16 codebook values (small, fits in registers)
+    # Use gather from codebook pointer
+    code_val = tl.load(codebook_ptr + nibble, mask=mask, other=0.0)
+
+    # Load absmax for this element's quantization block
+    block_idx = elem_offset // blocksize
+    absmax_global_offset = expert_global * blocks_per_expert + block_idx
+    absmax_val = tl.load(absmax_ptr + absmax_global_offset, mask=mask, other=1.0)
+
+    # Dequantize: value = codebook[nibble] * absmax
+    result = code_val * absmax_val
+
+    # Store to output
+    out_offset = expert_local_idx * stride_out_e + elem_offset
+    tl.store(out_ptr + out_offset, result.to(out_ptr.dtype.element_ty), mask=mask)
+
+
+def selective_dequant_nf4_triton(
+    packed_data: torch.Tensor,
+    absmax: torch.Tensor,
+    active_experts: torch.Tensor,
+    expert_shape: tuple[int, int],
+    blocksize: int,
+    dtype: torch.dtype = torch.bfloat16,
+    codebook: torch.Tensor | None = None,
+) -> torch.Tensor:
+    """Fused selective gather + NF4 dequantization via Triton kernel.
+
+    Args:
+        packed_data: Flattened packed NF4 data [total_packed] or [total_packed, 1]
+        absmax: Per-block scaling factors [total_blocks]
+        active_experts: Sorted indices of experts to dequantize [num_active]
+        expert_shape: (dim1, dim2) per expert
+        blocksize: Quantization block size
+        dtype: Output dtype (default bf16)
+        codebook: NF4 lookup table [16] (uses default NF4 codebook if None)
+
+    Returns:
+        Dequantized weights [num_active, dim1, dim2]
+    """
+    num_active = active_experts.shape[0]
+    expert_numel = expert_shape[0] * expert_shape[1]
+    packed_per_expert = expert_numel // 2
+    blocks_per_expert = expert_numel // blocksize
+
+    # Prepare codebook on device
+    if codebook is None:
+        codebook = torch.tensor(
+            NF4_CODEBOOK, dtype=torch.float32, device=packed_data.device
+        )
+    else:
+        codebook = codebook.to(device=packed_data.device, dtype=torch.float32)
+
+    # Flatten inputs
+    packed_flat = packed_data.reshape(-1)
+    absmax_flat = absmax.reshape(-1).float()  # absmax is usually fp32
+
+    # Output buffer
+    out = torch.empty(num_active, expert_numel, dtype=dtype, device=packed_data.device)
+
+    BLOCK_SIZE = 1024  # Process 1024 elements per thread block
+
+    grid = (num_active, triton.cdiv(expert_numel, BLOCK_SIZE))
+
+    _selective_dequant_nf4_kernel[grid](
+        packed_flat,
+        absmax_flat,
+        active_experts,
+        codebook,
+        out,
+        out.stride(0),
+        num_active=num_active,
+        packed_per_expert=packed_per_expert,
+        blocks_per_expert=blocks_per_expert,
+        blocksize=blocksize,
+        BLOCK_SIZE=BLOCK_SIZE,
+    )
+
+    return out.reshape(num_active, *expert_shape)

src/axolotl/integrations/kernels/plugin.py

@@ -61,7 +61,16 @@ class KernelsPlugin(BasePlugin):
         return "axolotl.integrations.kernels.KernelsArgs"
 
     def pre_model_load(self, cfg):
+        from axolotl.integrations.kernels.constants import SPARSE_MOE_BLOCK
+
+        # Prefer text backbone type for VLMs, but fall back to base type
+        # when the text type isn't in the supported mapping (e.g. qwen3_5_moe_text)
         moe_model_type = cfg.model_config_type_text or cfg.model_config_type
+        if (
+            moe_model_type not in SPARSE_MOE_BLOCK
+            and cfg.model_config_type in SPARSE_MOE_BLOCK
+        ):
+            moe_model_type = cfg.model_config_type
 
         if cfg.use_scattermoe:
             self._register_kernels()

src/axolotl/kernels/lora.py

@@ -640,7 +640,9 @@ class LoRA_QKV(torch.autograd.Function):
         del q_weight
         del q_weight_t
         if A_q is not None and B_q is not None:
-            grad_X.addmm_(q_grad, torch.mm(B_q_scaled, A_q_scaled))
+            # Stay decomposed: dQ @ B^T gives [T, R], then [T, R] @ (s*A) gives [T, in]
+            # This is 65x fewer FLOPs than materializing B@A into [out, in]
+            grad_X.addmm_(torch.mm(q_grad, B_q_scaled), A_q_scaled)
 
         # K path
         k_weight_t = dequantize(k_weight, k_quant)
@@ -648,7 +650,7 @@ class LoRA_QKV(torch.autograd.Function):
         del k_weight
         del k_weight_t
         if A_k is not None and B_k is not None:
-            grad_X.addmm_(k_grad, torch.mm(B_k_scaled, A_k_scaled))
+            grad_X.addmm_(torch.mm(k_grad, B_k_scaled), A_k_scaled)
 
         # V path
         v_weight_t = dequantize(v_weight, v_quant)
@@ -656,7 +658,7 @@ class LoRA_QKV(torch.autograd.Function):
         del v_weight
         del v_weight_t
         if A_v is not None and B_v is not None:
-            grad_X.addmm_(v_grad, torch.mm(B_v_scaled, A_v_scaled))
+            grad_X.addmm_(torch.mm(v_grad, B_v_scaled), A_v_scaled)
 
         # Transpose gradients if needed
         if d_A_q is not None:
@@ -819,7 +821,8 @@ class LoRA_O(torch.autograd.Function):
         del W
 
         A, B = A.to(dtype), B.to(dtype)
-        dX += s * dY @ B @ A
+        # Stay decomposed: dY @ B gives [T, R], then [T, R] @ A gives [T, in]
+        dX.addmm_(torch.mm(dY, B), A, alpha=s)
 
         # W, b, W_quant, A, B, s
         return dX.view(batch, seq_len, hd), None, None, None, d_A.t(), d_B.t(), None

src/axolotl/loaders/model.py

@@ -505,6 +505,20 @@ class ModelLoader:
         elif not is_ds_zero3:
             self.model_kwargs["device_map"] = device_map
 
+            # quantize_moe_experts quantizes expert weights on-the-fly during loading,
+            # so the actual VRAM usage is much less than bf16 estimates.
+            # When device_map is "auto", accelerate's infer_auto_device_map computes
+            # the device map at bf16 size (before quantization), causing it to offload
+            # layers to CPU, which BnB then rejects. Force single-GPU placement to
+            # prevent this. Only applies to the non-FSDP, non-ZeRO3 path (DDP/single).
+            if getattr(self.cfg, "quantize_moe_experts", False) and device_map in (
+                "auto",
+                None,
+            ):
+                self.model_kwargs["device_map"] = {
+                    "": int(os.environ.get("LOCAL_RANK", 0))
+                }
+
             cur_device = get_device_type()
             if "mps" in str(cur_device):
                 self.model_kwargs["device_map"] = "mps:0"

src/axolotl/loaders/patch_manager.py

@@ -571,15 +571,6 @@ class PatchManager:
         LOG.info("Patching with xformers attention...")
         hijack_llama_attention()
 
-    def _patch_llama_sample_packing(self):
-        """Apply sample packing patches for LLaMA models."""
-        from axolotl.monkeypatch.llama_patch_multipack import (
-            hijack_llama_prepare_4d_mask,
-        )
-
-        LOG.info("Patching llama _prepare_4d_causal_attention_mask*...")
-        hijack_llama_prepare_4d_mask()
-
     def _patch_llama_derived_model(self):
         """Modify all llama derived models in one block."""
         if self.cfg.is_llama_derived_model and not (
@@ -591,8 +582,6 @@ class PatchManager:
                 self._patch_llama_flash_attention()
             elif self.cfg.xformers_attention:
                 self._patch_llama_xformers_attention()
-            elif self.cfg.sample_packing:
-                self._patch_llama_sample_packing()
             elif self.cfg.s2_attention:
                 raise NotImplementedError(
                     "Shifted-sparse attention not currently implemented without flash attention."

src/axolotl/loaders/tokenizer.py

@@ -221,6 +221,14 @@ def load_tokenizer(cfg: DictDefault) -> PreTrainedTokenizer:
             if getattr(tokenizer, attr_name) is None:
                 setattr(tokenizer, attr_name, "<|endoftext|>")
 
+    # Generic fallback: if tokenizer still has no pad_token, use eos_token
+    if tokenizer.pad_token is None and tokenizer.eos_token is not None:
+        tokenizer.pad_token = tokenizer.eos_token
+        LOG.warning(
+            "Tokenizer does not have a pad_token, falling back to eos_token: %s",
+            tokenizer.eos_token,
+        )
+
     additional_special_tokens = None
     if cfg.special_tokens:
         special_tokens = cfg.special_tokens.to_dict()

src/axolotl/monkeypatch/accelerate/parallelism_config.py

@@ -78,30 +78,21 @@ def patch_parallelism_config():
 
 
 def patch_prepare_cp():
-    import functools
+    import contextlib
 
-    import torch
     from accelerate import Accelerator
 
     def patched_prepare_cp(self, *args):
         if self.parallelism_config.cp_backend == "deepspeed":
             return args
 
-        from accelerate.big_modeling import _attach_context_parallel_hooks
-        from torch.distributed.tensor.experimental import context_parallel
-        from torch.distributed.tensor.experimental._attention import set_rotate_method
-
-        cp_comm_strategy = self.parallelism_config.cp_handler.cp_comm_strategy
-        set_rotate_method(cp_comm_strategy)
-
-        self._cp_context = functools.partial(
-            context_parallel, mesh=self.torch_device_mesh["cp"]
-        )
-
-        for arg in args:
-            if isinstance(arg, torch.nn.Module):
-                _attach_context_parallel_hooks(arg)
+        @contextlib.contextmanager
+        def _noop_cp_context(
+            buffers=None, buffer_seq_dims=None, no_restore_buffers=None
+        ):
+            yield
 
+        self._cp_context = _noop_cp_context
         return args
 
     Accelerator._prepare_cp = patched_prepare_cp

src/axolotl/monkeypatch/llama_expand_mask.py

@@ -1,24 +0,0 @@
-"""
-expands the binary attention mask per 3.2.2 of https://arxiv.org/pdf/2107.02027.pdf
-"""
-
-from typing import Optional
-
-import torch
-
-from axolotl.monkeypatch.utils import mask_2d_to_4d
-
-
-def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
-    masked_zero_one_mask = mask_2d_to_4d(mask, dtype, tgt_len)
-    inverted_mask = 1.0 - masked_zero_one_mask
-
-    return inverted_mask.masked_fill(
-        inverted_mask.to(torch.bool), torch.finfo(dtype).min
-    )
-
-
-def hijack_expand_mask():
-    import transformers
-
-    transformers.models.llama.modeling_llama._expand_mask = _expand_mask

src/axolotl/monkeypatch/llama_patch_multipack.py

@@ -1,26 +0,0 @@
-"""
-Patched LlamaAttention to use torch.nn.functional.scaled_dot_product_attention
-"""
-
-from axolotl.monkeypatch.utils import (
-    patched_prepare_4d_causal_attention_mask,
-    patched_prepare_4d_causal_attention_mask_for_sdpa,
-)
-
-
-def hijack_llama_prepare_4d_mask():
-    from transformers import modeling_attn_mask_utils
-    from transformers.models.llama import modeling_llama
-
-    modeling_llama._prepare_4d_causal_attention_mask_for_sdpa = (
-        patched_prepare_4d_causal_attention_mask_for_sdpa
-    )
-    modeling_attn_mask_utils._prepare_4d_causal_attention_mask_for_sdpa = (
-        patched_prepare_4d_causal_attention_mask_for_sdpa
-    )
-    modeling_llama._prepare_4d_causal_attention_mask = (
-        patched_prepare_4d_causal_attention_mask
-    )
-    modeling_attn_mask_utils._prepare_4d_causal_attention_mask = (
-        patched_prepare_4d_causal_attention_mask
-    )

src/axolotl/monkeypatch/lora_kernels.py

@@ -51,6 +51,29 @@ QKV_PATCHES = [
     value_states = value_states.view(hidden_shape).transpose(1, 2)
 """.lstrip("\n"),
     ),
+    (
+        """
+    query_states, gate = torch.chunk(
+        self.q_proj(hidden_states).view(*input_shape, -1, self.head_dim * 2), 2, dim=-1
+    )
+    gate = gate.reshape(*input_shape, -1)
+
+    query_states = self.q_norm(query_states.view(hidden_shape)).transpose(1, 2)
+    key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+""".lstrip("\n"),
+        """
+    query_states, key_states, value_states = self.apply_qkv(hidden_states)
+    query_states, gate = torch.chunk(
+        query_states.view(*input_shape, -1, self.head_dim * 2), 2, dim=-1
+    )
+    gate = gate.reshape(*input_shape, -1)
+
+    query_states = self.q_norm(query_states.view(hidden_shape)).transpose(1, 2)
+    key_states = self.k_norm(key_states.view(hidden_shape)).transpose(1, 2)
+    value_states = value_states.view(hidden_shape).transpose(1, 2)
+""".lstrip("\n"),
+    ),
 ]
 
 ORIGINAL_O_CODE = """
@@ -299,6 +322,8 @@ def get_layers(model: PeftModelForCausalLM) -> list[nn.Module]:
     if hasattr(pretrained_model, "language_model"):
         return pretrained_model.language_model.layers
     if hasattr(pretrained_model, "model"):
+        if hasattr(pretrained_model.model, "language_model"):
+            return pretrained_model.model.language_model.layers
         return pretrained_model.model.layers
 
     raise NotImplementedError(

src/axolotl/monkeypatch/multipack.py

@@ -59,6 +59,7 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "ministral3",
     "mistral4",
     "afmoe",
+    "nemotron",
 ]
 
 

src/axolotl/monkeypatch/utils.py

@@ -3,15 +3,10 @@ Shared utils for the monkeypatches
 """
 
 import re
-from typing import Optional, Tuple
+from typing import Tuple
 
 import torch
 import torch.nn.functional as F
-from transformers.modeling_attn_mask_utils import (
-    _prepare_4d_causal_attention_mask,
-    _prepare_4d_causal_attention_mask_for_sdpa,
-)
-from transformers.utils import is_torch_bf16_gpu_available
 
 
 @torch.jit.script
@@ -170,65 +165,6 @@ def set_module_name(model, name, value):
     setattr(parent, child_name, value)
 
 
-def mask_2d_to_4d(
-    mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None
-):
-    """
-    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
-    This expansion handles packed sequences so that sequences share the same attention mask integer value
-    when they attend to each other within that sequence.
-    This expansion transforms the mask to lower triangular form to prevent future peeking.
-    """
-    bsz, src_len = mask.size()
-    tgt_len = tgt_len if tgt_len is not None else src_len
-
-    mask = mask.unsqueeze(1).unsqueeze(2)
-    mask = mask.expand(bsz, 1, tgt_len, src_len)
-
-    # Create a binary mask from the original mask where zeros remain zeros and all other values are set to one
-    binary_mask = torch.where(
-        mask != 0,
-        torch.tensor(1, device=mask.device).to(dtype),
-        torch.tensor(0, device=mask.device).to(dtype),
-    )
-
-    # Create a block-diagonal mask.
-    # we multiply by the binary mask so that 0's in the original mask are correctly excluded
-    zero_one_mask = torch.eq(mask, mask.transpose(-1, -2)).int() * binary_mask
-
-    # Now let's create a lower triangular mask of ones that will zero out the upper triangular part
-    lower_triangular_ones = torch.tril(torch.ones((tgt_len, src_len), dtype=dtype)).to(
-        mask.device
-    )
-
-    # Use the lower triangular mask to zero out the upper triangular part of the zero_one_mask
-    masked_zero_one_mask = zero_one_mask * lower_triangular_ones
-
-    return masked_zero_one_mask
-
-
-def patched_prepare_4d_causal_attention_mask(
-    attention_mask: Optional[torch.Tensor],
-    *args,
-):
-    dtype = torch.bfloat16 if is_torch_bf16_gpu_available() else torch.float32
-    return _prepare_4d_causal_attention_mask(
-        mask_2d_to_4d(attention_mask, dtype=dtype),
-        *args,
-    )
-
-
-def patched_prepare_4d_causal_attention_mask_for_sdpa(
-    attention_mask: Optional[torch.Tensor],
-    *args,
-):
-    dtype = torch.bfloat16 if is_torch_bf16_gpu_available() else torch.float32
-    return _prepare_4d_causal_attention_mask_for_sdpa(
-        mask_2d_to_4d(attention_mask, dtype=dtype),
-        *args,
-    )
-
-
 def detab_code(code: str) -> Tuple[str, str]:
     try:
         spaces = re.match(r"([\s\t]{1,})", code).group(0)

src/axolotl/prompt_strategies/_synthetic.py

@@ -0,0 +1,96 @@
+"""
+Synthetic dataset generator for benchmarking and testing.
+
+Generates datasets with configurable sequence length, dataset size, and token ID ranges.
+Useful for benchmarking memory usage and speed by sequence length, and for validating
+weighted dataset mixes.
+
+YAML configuration example:
+
+    datasets:
+      - path: synthetic
+        type: _synthetic
+        length: 1000
+        sequence_length: 2048
+        min_input_id: 100
+        max_input_id: 32000
+        seed: 42
+"""
+
+from typing import Any, Dict, Optional
+
+import numpy as np
+from datasets import Dataset
+
+from axolotl.prompt_tokenizers import DatasetWrappingStrategy
+from axolotl.utils.logging import get_logger
+
+LOG = get_logger(__name__)
+
+
+class SyntheticDatasetStrategy(DatasetWrappingStrategy):
+    """Strategy that generates synthetic tokenized data, ignoring the source dataset."""
+
+    def __init__(
+        self,
+        sequence_length: int = 2048,
+        length: int = 1000,
+        min_input_id: int = 100,
+        max_input_id: int = 32000,
+        seed: Optional[int] = None,
+    ):
+        self.sequence_length = sequence_length
+        self.length = length
+        self.min_input_id = min_input_id
+        self.max_input_id = max_input_id
+        self.seed = seed
+
+    def wrap_dataset(
+        self,
+        dataset,
+        process_count: int | None = None,
+        keep_in_memory: bool | None = False,
+        **kwargs,
+    ) -> Dataset:
+        LOG.info(
+            f"Generating synthetic dataset: {self.length} samples, "
+            f"sequence_length={self.sequence_length}, "
+            f"input_id_range=[{self.min_input_id}, {self.max_input_id})"
+        )
+
+        rng = np.random.default_rng(self.seed)
+        input_ids = rng.integers(
+            low=self.min_input_id,
+            high=self.max_input_id,
+            size=(self.length, self.sequence_length),
+        ).tolist()
+
+        attention_mask = [[1] * self.sequence_length] * self.length
+        # labels == input_ids means we train on all tokens
+        labels = [row[:] for row in input_ids]
+
+        return Dataset.from_dict(
+            {
+                "input_ids": input_ids,
+                "attention_mask": attention_mask,
+                "labels": labels,
+            }
+        )
+
+
+def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
+    ds_cfg = ds_cfg or {}
+
+    sequence_length = ds_cfg.get("sequence_length", cfg.sequence_len)
+    length = ds_cfg.get("length", 1000)
+    min_input_id = ds_cfg.get("min_input_id", 100)
+    max_input_id = ds_cfg.get("max_input_id", tokenizer.vocab_size)
+    seed = ds_cfg.get("seed", None)
+
+    return SyntheticDatasetStrategy(
+        sequence_length=sequence_length,
+        length=length,
+        min_input_id=min_input_id,
+        max_input_id=max_input_id,
+        seed=seed,
+    )

src/axolotl/train.py

@@ -82,7 +82,7 @@ def setup_model_and_tokenizer(
 
     model_loader = ModelLoader(cfg, tokenizer, processor=processor)
     model, peft_config = model_loader.load()
-    if model.generation_config is not None:
+    if getattr(model, "generation_config", None) is not None:
         model.generation_config.do_sample = True
 
     model_properties = model.config.to_dict()

src/axolotl/utils/callbacks/profiler.py

@@ -17,6 +17,8 @@ from transformers import (
 class PytorchProfilerCallback(TrainerCallback):
     """
     PyTorch Profiler callback to create snapshots of GPU memory usage at specified steps.
+
+    Also runs torch.profiler to produce a Chrome trace for timing analysis.
     """
 
     def __init__(self, steps_to_profile: int = 5, profiler_steps_start: int = 0):
@@ -26,9 +28,10 @@ class PytorchProfilerCallback(TrainerCallback):
         if profiler_steps_start == 0:
             # start recording memory allocations before everything is allocated, because if we start
             # at the beginning of step 0, we won't have any memory allocations in the traces
-            torch.cuda.memory._record_memory_history(enabled="all")
+            torch.cuda.memory._record_memory_history(enabled="all", stacks="all")
             profiler_steps_start = -1
         self.profiler_steps_start = profiler_steps_start
+        self._profiler = None
 
     def on_step_begin(
         self,
@@ -38,7 +41,21 @@ class PytorchProfilerCallback(TrainerCallback):
         **kwargs,
     ):
         if state.global_step == self.profiler_steps_start:
-            torch.cuda.memory._record_memory_history(enabled="all")
+            torch.cuda.memory._record_memory_history(enabled="all", stacks="all")
+
+        # Start torch.profiler on the first profiled step
+        if state.global_step == max(self.profiler_steps_start, 0):
+            profiler = torch.profiler.profile(
+                activities=[
+                    torch.profiler.ProfilerActivity.CPU,
+                    torch.profiler.ProfilerActivity.CUDA,
+                ],
+                record_shapes=True,
+                profile_memory=True,
+                with_stack=True,
+            )
+            profiler.__enter__()
+            self._profiler = profiler
 
     def on_step_end(
         self,
@@ -55,6 +72,13 @@ class PytorchProfilerCallback(TrainerCallback):
             # tell CUDA to stop recording memory allocations now
             torch.cuda.memory._record_memory_history(enabled=None)
 
+            # Stop and export torch.profiler trace
+            if self._profiler is not None:
+                self._profiler.__exit__(None, None, None)
+                trace_path = Path(args.output_dir) / "profiler_trace.json"
+                self._profiler.export_chrome_trace(str(trace_path))
+                self._profiler = None
+
     def on_train_end(
         self,
         args: TrainingArguments,
@@ -73,3 +97,9 @@ class PytorchProfilerCallback(TrainerCallback):
 
             # tell CUDA to stop recording memory allocations now
             torch.cuda.memory._record_memory_history(enabled=None)
+
+        if self._profiler is not None:
+            self._profiler.__exit__(None, None, None)
+            trace_path = Path(args.output_dir) / "profiler_trace.json"
+            self._profiler.export_chrome_trace(str(trace_path))
+            self._profiler = None

src/axolotl/utils/callbacks/qat.py

@@ -25,9 +25,11 @@ def toggle_fake_quant(mod: nn.Module, enable: bool):
         if (
             isinstance(mod, FakeQuantizedLinear)
             and mod.activation_fake_quantizer is not None
+            and hasattr(mod.activation_fake_quantizer, "enabled")
         ):
             mod.activation_fake_quantizer.enabled = enable
-        mod.weight_fake_quantizer.enabled = enable
+        if hasattr(mod.weight_fake_quantizer, "enabled"):
+            mod.weight_fake_quantizer.enabled = enable
 
 
 class QATCallback(TrainerCallback):

src/axolotl/utils/callbacks/tokens_per_second.py

@@ -12,12 +12,11 @@ from transformers import (
     TrainingArguments,
 )
 
+from axolotl.core.trainers.constants import TOKENS_STATE_FILE
 from axolotl.utils.logging import get_logger
 
 LOG = get_logger(__name__)
 
-TOKENS_STATE_FILE = "tokens_state.json"
-
 
 class TokensPerSecondCallback(TrainerCallback):
     """

src/axolotl/utils/config/__init__.py

@@ -22,7 +22,12 @@ from axolotl.utils.schemas.config import (
     AxolotlConfigWCapabilities as AxolotlConfigWCapabilitiesBase,
     AxolotlInputConfig as AxolotlInputConfigBase,
 )
-from axolotl.utils.schemas.datasets import DPODataset, KTODataset, SFTDataset
+from axolotl.utils.schemas.datasets import (
+    DPODataset,
+    KTODataset,
+    SFTDataset,
+    SyntheticDataset,
+)
 
 LOG = get_logger(__name__)
 
@@ -308,6 +313,14 @@ def validate_config(
                 cfg["datasets"][idx] = DPODataset(**ds_cfg)
             elif cfg.get("rl") == "kto" and not isinstance(ds_cfg, KTODataset):
                 cfg["datasets"][idx] = KTODataset(**dict(ds_cfg))
+            elif (
+                ds_cfg.get("type")
+                if isinstance(ds_cfg, dict)
+                else getattr(ds_cfg, "type", None)
+            ) == "_synthetic" and not isinstance(ds_cfg, SyntheticDataset):
+                cfg["datasets"][idx] = SyntheticDataset(
+                    **(ds_cfg if isinstance(ds_cfg, dict) else dict(ds_cfg))
+                )
             elif not isinstance(ds_cfg, SFTDataset):
                 cfg["datasets"][idx] = SFTDataset(**dict(ds_cfg))
 

src/axolotl/utils/data/sft.py

@@ -376,10 +376,14 @@ def _load_and_process_single_dataset(
     streaming: bool = False,
 ) -> tuple[Dataset | IterableDataset, Prompter | None]:
     """Load and process a single dataset based on the passed config."""
-    # Load the dataset
-    dataset = load_dataset_with_config(
-        dataset_config, cfg.hf_use_auth_token, streaming=streaming
-    )
+    # For synthetic datasets, create a minimal placeholder instead of loading from path
+    if dataset_config.type == "_synthetic":
+        dataset = Dataset.from_dict({"text": [""]})
+    else:
+        # Load the dataset
+        dataset = load_dataset_with_config(
+            dataset_config, cfg.hf_use_auth_token, streaming=streaming
+        )
 
     # Parse dataset type
     d_base_type, d_prompt_style = _parse_dataset_type(dataset_config.type)

src/axolotl/utils/quantization.py

@@ -10,9 +10,11 @@ from torchao.quantization import quantize_
 from torchao.quantization.qat import (
     QATConfig,
 )
+from torchao.quantization.qat.fake_quantize_config import Int4WeightFakeQuantizeConfig
 from torchao.quantization.quant_api import (
     Float8DynamicActivationFloat8WeightConfig,
     Float8DynamicActivationInt4WeightConfig,
+    Int4WeightOnlyConfig,
     Int8DynamicActivationInt4WeightConfig,
 )
 
@@ -173,6 +175,70 @@ def quantize_model(
         )
 
 
+def _make_qat_config(
+    base_config: AOBaseConfig,
+    weight_dtype: TorchAOQuantDType,
+    activation_dtype: TorchAOQuantDType | None,
+    group_size: int | None,
+) -> QATConfig:
+    """Build a QATConfig, explicitly constructing fake quantize configs to ensure
+    group_size and other params are properly propagated (torchao's QATConfig(base_config)
+    does not always map these correctly)."""
+    from torchao.quantization.qat.fake_quantize_config import (
+        Float8FakeQuantizeConfig,
+        IntxFakeQuantizeConfig,
+    )
+
+    if isinstance(base_config, MXFakeQuantizeConfig):
+        return QATConfig(
+            activation_config=base_config,
+            weight_config=base_config,
+        )
+
+    # Build explicit weight config
+    weight_fq_config: (
+        Int4WeightFakeQuantizeConfig
+        | IntxFakeQuantizeConfig
+        | Float8FakeQuantizeConfig
+        | None
+    ) = None
+    if weight_dtype == TorchAOQuantDType.int4:
+        gs = (
+            group_size
+            if group_size is not None
+            else getattr(base_config, "group_size", 128)
+        )
+        activation_dt = None
+        if activation_dtype == TorchAOQuantDType.int8:
+            activation_dt = torch.bfloat16
+        elif activation_dtype == TorchAOQuantDType.float8_e4m3fn:
+            activation_dt = torch.float8_e4m3fn
+        kwargs = {"group_size": gs}
+        if activation_dt is not None:
+            kwargs["activation_dtype"] = activation_dt
+        weight_fq_config = Int4WeightFakeQuantizeConfig(**kwargs)
+    elif weight_dtype == TorchAOQuantDType.float8_e4m3fn:
+        weight_fq_config = Float8FakeQuantizeConfig(dtype=torch.float8_e4m3fn)
+
+    # Build explicit activation config
+    activation_fq_config = None
+    if activation_dtype == TorchAOQuantDType.int8:
+        activation_fq_config = IntxFakeQuantizeConfig(
+            dtype=torch.int8, granularity="per_token", is_symmetric=False
+        )
+    elif activation_dtype == TorchAOQuantDType.float8_e4m3fn:
+        activation_fq_config = Float8FakeQuantizeConfig(dtype=torch.float8_e4m3fn)
+
+    if weight_fq_config is not None:
+        return QATConfig(
+            weight_config=weight_fq_config,
+            activation_config=activation_fq_config,
+        )
+
+    # Fallback to base_config for unhandled combos
+    return QATConfig(base_config)
+
+
 def prepare_model_for_qat(
     model,
     weight_dtype: TorchAOQuantDType,
@@ -200,13 +266,9 @@ def prepare_model_for_qat(
         activation_dtype=activation_dtype,
         group_size=group_size,
     )
-    if isinstance(base_config, MXFakeQuantizeConfig):
-        qat_config = QATConfig(
-            activation_config=base_config,
-            weight_config=base_config,
-        )
-    else:
-        qat_config = QATConfig(base_config)
+    qat_config = _make_qat_config(
+        base_config, weight_dtype, activation_dtype, group_size
+    )
     quantize_(model, qat_config)
     if quantize_embedding:
         # activation fake quantization is not supported for embedding layers
@@ -215,12 +277,9 @@ def prepare_model_for_qat(
             activation_dtype=None,
             group_size=group_size,
         )
-        if isinstance(embedding_base_config, MXFakeQuantizeConfig):
-            embedding_qat_config = QATConfig(
-                weight_config=embedding_base_config,
-            )
-        else:
-            embedding_qat_config = QATConfig(embedding_base_config)
+        embedding_qat_config = _make_qat_config(
+            embedding_base_config, weight_dtype, None, group_size
+        )
         quantize_(
             model,
             embedding_qat_config,

src/axolotl/utils/schedulers.py

@@ -2,7 +2,7 @@
 
 import math
 from functools import partial
-from typing import Sequence
+from typing import Any, Sequence
 
 from torch import Tensor
 from torch.optim import Optimizer
@@ -340,3 +340,19 @@ class JaggedLRRestartScheduler(LRScheduler):
             return [lr * scale for lr in original]
 
         return original * scale
+
+    def state_dict(self) -> dict[str, Any]:
+        """Return serializable state, saving inner_schedule as its own state_dict."""
+        state = {
+            key: value
+            for key, value in self.__dict__.items()
+            if key not in ("optimizer", "inner_schedule")
+        }
+        state["inner_schedule_state"] = self.inner_schedule.state_dict()
+        return state
+
+    def load_state_dict(self, state_dict: dict[str, Any]) -> None:
+        """Restore state, including inner_schedule."""
+        inner_state = state_dict.pop("inner_schedule_state")
+        self.__dict__.update(state_dict)
+        self.inner_schedule.load_state_dict(inner_state)

src/axolotl/utils/schemas/config.py

@@ -22,6 +22,7 @@ from axolotl.utils.schemas.datasets import (
     PretrainingDataset,
     SFTDataset,
     StepwiseSupervisedDataset,
+    SyntheticDataset,
 )
 from axolotl.utils.schemas.deprecated import DeprecatedParameters, RemappedParameters
 from axolotl.utils.schemas.dynamic_checkpoint import DynamicCheckpointConfig
@@ -185,7 +186,13 @@ class AxolotlInputConfig(
 
     datasets: (
         Annotated[
-            list[SFTDataset | DPODataset | KTODataset | StepwiseSupervisedDataset],
+            list[
+                SFTDataset
+                | DPODataset
+                | KTODataset
+                | StepwiseSupervisedDataset
+                | SyntheticDataset
+            ],
             MinLen(1),
         ]
         | None
@@ -198,7 +205,13 @@ class AxolotlInputConfig(
 
     test_datasets: (
         Annotated[
-            list[SFTDataset | DPODataset | KTODataset | StepwiseSupervisedDataset],
+            list[
+                SFTDataset
+                | DPODataset
+                | KTODataset
+                | StepwiseSupervisedDataset
+                | SyntheticDataset
+            ],
             MinLen(1),
         ]
         | None
@@ -433,6 +446,12 @@ class AxolotlInputConfig(
             "description": "Whether to offload activations. Available options are: true, false, 'legacy', 'disk'."
         },
     )
+    layer_offloading: bool | None = Field(
+        default=False,
+        json_schema_extra={
+            "description": "Offload model layer parameters to CPU during forward, prefetch back during backward."
+        },
+    )
 
     unfrozen_parameters: list[str] | None = Field(
         default=None,

src/axolotl/utils/schemas/datasets.py

@@ -296,4 +296,42 @@ class KTODataset(BaseModel):
     revision: str | None = None
 
 
-DatasetConfig = SFTDataset | DPODataset | KTODataset | StepwiseSupervisedDataset
+class SyntheticDataset(BaseModel):
+    """Synthetic dataset configuration for benchmarking and testing.
+
+    Generates datasets with configurable sequence length, dataset size, and token ID
+    ranges. Useful for benchmarking memory usage and speed by sequence length, and for
+    validating weighted dataset mixes.
+    """
+
+    path: Literal["synthetic"] = "synthetic"
+    type: Literal["_synthetic"] = "_synthetic"
+    length: int = Field(
+        default=1000,
+        json_schema_extra={"description": "Number of rows to generate"},
+    )
+    sequence_length: int | None = Field(
+        default=None,
+        json_schema_extra={
+            "description": "Sequence length per row (defaults to sequence_len from config)"
+        },
+    )
+    min_input_id: int = Field(
+        default=100,
+        json_schema_extra={"description": "Minimum token ID for generation"},
+    )
+    max_input_id: int | None = Field(
+        default=None,
+        json_schema_extra={
+            "description": "Maximum token ID for generation (defaults to tokenizer vocab_size)"
+        },
+    )
+    seed: int | None = Field(
+        default=None,
+        json_schema_extra={"description": "Random seed for reproducibility"},
+    )
+
+
+DatasetConfig = (
+    SFTDataset | DPODataset | KTODataset | StepwiseSupervisedDataset | SyntheticDataset
+)

src/axolotl/utils/schemas/enums.py

@@ -87,6 +87,11 @@ class CustomSupportedOptimizers(str, Enum):
     came_pytorch = "came_pytorch"
     muon = "muon"
     dion = "dion"
+    flash_adamw = "flash_adamw"
+    flash_adam = "flash_adam"
+    flash_sgd = "flash_sgd"
+    flash_sgdw = "flash_sgdw"
+    flash_lion = "flash_lion"
 
 
 class RingAttnFunc(str, Enum):

src/axolotl/utils/schemas/validation.py

@@ -253,6 +253,23 @@ class TrainingValidationMixin:
                 data["pad_to_sequence_len"] = True
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def set_reward_model_defaults(cls, data):
+        if data.get("reward_model"):
+            if data.get("num_labels") is None:
+                data["num_labels"] = 1
+            if not (data.get("type_of_model") or data.get("model_type")):
+                data["model_type"] = "AutoModelForSequenceClassification"
+
+        if data.get("process_reward_model"):
+            if data.get("num_labels") is None:
+                data["num_labels"] = 2
+            if not (data.get("type_of_model") or data.get("model_type")):
+                data["model_type"] = "AutoModelForTokenClassification"
+
+        return data
+
     @model_validator(mode="before")
     @classmethod
     def check_gas_bsz(cls, data):
@@ -773,6 +790,14 @@ class OptimizationValidationMixin:
             LOG.warning("adamw hyperparameters found, but no adamw optimizer set")
         return self
 
+    @staticmethod
+    def _resolve_fsdp_version(data):
+        """Resolve FSDP version from top-level fsdp_version or fsdp_config.fsdp_version."""
+        fsdp_version = data.get("fsdp_version")
+        if fsdp_version is None:
+            fsdp_version = data.get("fsdp_config", {}).get("fsdp_version", 1)
+        return fsdp_version
+
     @model_validator(mode="before")
     @classmethod
     def check_muon_deepspeed_fsdp(cls, data):
@@ -782,15 +807,32 @@ class OptimizationValidationMixin:
                     "Muon optimizer is currently incompatible with DeepSpeed"
                 )
             if data.get("fsdp") or data.get("fsdp_config"):
-                fsdp_version = data.get("fsdp_version")
-                if fsdp_version is None:
-                    fsdp_version = data.get("fsdp_config", {}).get("fsdp_version", 1)
+                fsdp_version = cls._resolve_fsdp_version(data)
                 if str(fsdp_version) != "2":
                     raise ValueError(
                         "Muon optimizer is only compatible with FSDP2. Set fsdp_version: 2 to use Muon with FSDP."
                     )
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_flashoptim_deepspeed_fsdp(cls, data):
+        optimizer = data.get("optimizer") or ""
+        if str(optimizer).startswith("flash_"):
+            if data.get("deepspeed"):
+                raise ValueError(
+                    f"{optimizer} optimizer is incompatible with DeepSpeed. "
+                    "Flash optimizers only support DDP and FSDP2."
+                )
+            if data.get("fsdp") or data.get("fsdp_config"):
+                fsdp_version = cls._resolve_fsdp_version(data)
+                if str(fsdp_version) != "2":
+                    raise ValueError(
+                        f"{optimizer} optimizer is only compatible with FSDP2. "
+                        "Set fsdp_version: 2 to use flash optimizers with FSDP."
+                    )
+        return data
+
     @model_validator(mode="before")
     @classmethod
     def check_batch_flattening_fa(cls, data):

tests/conftest.py

@@ -15,6 +15,8 @@ import datasets
 import pytest
 import requests
 import torch
+import transformers.utils as _transformers_utils
+import transformers.utils.import_utils as _import_utils
 from huggingface_hub import snapshot_download
 from huggingface_hub.errors import LocalEntryNotFoundError
 from tokenizers import AddedToken
@@ -29,6 +31,26 @@ from tests.hf_offline_utils import (
 
 logging.getLogger("filelock").setLevel(logging.CRITICAL)
 
+# Shim for deepseek v3
+if not hasattr(_import_utils, "is_torch_fx_available"):
+
+    def _is_torch_fx_available():
+        try:
+            import torch.fx  # noqa: F401  # pylint: disable=unused-import
+
+            return True
+        except ImportError:
+            return False
+
+    _import_utils.is_torch_fx_available = _is_torch_fx_available
+
+if not hasattr(_transformers_utils, "is_flash_attn_greater_or_equal_2_10"):
+    from transformers.utils import is_flash_attn_greater_or_equal as _is_flash_attn_gte
+
+    _transformers_utils.is_flash_attn_greater_or_equal_2_10 = lambda: (
+        _is_flash_attn_gte("2.10")
+    )
+
 
 def retry_on_request_exceptions(max_retries=3, delay=1):
     def decorator(func):

tests/core/test_builders.py

@@ -536,7 +536,7 @@ class TestHFCausalTrainerBuilder:
         "cfg_string",
         [
             "sft_cfg",
-            # "rm_cfg",  # TODO fix for num_labels = 2 vs 1
+            "rm_cfg",
             "prm_cfg",
         ],
     )

tests/e2e/integrations/test_scattermoe_lora_kernels.py

@@ -20,6 +20,7 @@ Test strategy:
 - Tolerances account for tf32 accumulation in Triton kernels
 """
 
+from functools import wraps
 from types import SimpleNamespace
 
 import pytest
@@ -34,6 +35,21 @@ pytestmark = pytest.mark.skipif(
 _SMOE = "axolotl.integrations.kernels.libs.scattermoe_lora"
 
 
+def skip_on_out_of_resources(func):
+    """Skip test if Triton kernel exceeds GPU shared memory limits."""
+
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        try:
+            return func(*args, **kwargs)
+        except Exception as exc:  # pylint: disable=broad-except
+            if "OutOfResources" in type(exc).__name__:
+                pytest.skip(f"GPU shared memory too small: {exc}")
+            raise
+
+    return wrapper
+
+
 # =============================================================================
 # Helpers
 # =============================================================================
@@ -209,6 +225,7 @@ def make_test_data(
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestForwardPass:
     """Test forward pass of fused scatter2scatter_lora kernel."""
 
@@ -288,6 +305,7 @@ class TestForwardPass:
         )
 
 
+@pytest.mark.slow
 class TestForwardGrouped:
     """Test forward pass with grouped_in/grouped_out configurations."""
 
@@ -377,6 +395,7 @@ class TestForwardGrouped:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestLoRAGradients:
     """Test backward LoRA gradient computation (dA, dB)."""
 
@@ -452,6 +471,7 @@ class TestLoRAGradients:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestAutograd:
     """Test full autograd integration through ScatterMoELoRA."""
 
@@ -620,6 +640,7 @@ class TestAutograd:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestBaseEquivalence:
     """When scaling=0, fused kernel should match base scatter2scatter."""
 
@@ -692,6 +713,7 @@ class TestBaseEquivalence:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestLoRAAdditivity:
     """Test that the LoRA component is correctly additive."""
 
@@ -749,6 +771,7 @@ class TestLoRAAdditivity:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestParallelExpertsModule:
     """Test the ParallelExperts module with LoRA."""
 
@@ -816,6 +839,7 @@ class TestParallelExpertsModule:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestEdgeCases:
     """Edge cases and boundary conditions."""
 
@@ -913,6 +937,7 @@ class TestEdgeCases:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestFusedDX:
     """Test fused backward dX kernel: dX = dY @ W^T + scaling * (dY @ B) @ A."""
 
@@ -980,6 +1005,7 @@ class TestFusedDX:
     def test_basic(self):
         self._run_fused_dX_test(M=32, K=64, N=128, E=4, R=8, k=2)
 
+    @skip_on_out_of_resources
     def test_large(self):
         self._run_fused_dX_test(M=256, K=256, N=512, E=8, R=16, k=2)
 
@@ -1122,6 +1148,7 @@ class TestFusedDX:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestFusedGatherBackward:
     """Test fused gather + backward dA/dB kernel."""
 
@@ -1174,6 +1201,7 @@ class TestFusedGatherBackward:
     def test_basic(self):
         self._run_fused_gather_test(M=32, K=64, N=128, E=4, R=8, k=2)
 
+    @skip_on_out_of_resources
     def test_large(self):
         self._run_fused_gather_test(M=256, K=256, N=512, E=8, R=16, k=2)
 
@@ -1183,6 +1211,7 @@ class TestFusedGatherBackward:
     def test_k1(self):
         self._run_fused_gather_test(M=64, K=64, N=128, E=4, R=8, k=1)
 
+    @skip_on_out_of_resources
     def test_many_experts(self):
         self._run_fused_gather_test(M=128, K=64, N=128, E=16, R=8, k=4)
 
@@ -1269,6 +1298,8 @@ class TestFusedGatherBackward:
 # =============================================================================
 
 
+@pytest.mark.slow
+@pytest.mark.xfail(reason="flaky", strict=False)
 class TestTokenRounding:
     """Test token rounding utility and its integration with backward kernels."""
 
@@ -1315,6 +1346,7 @@ class TestTokenRounding:
                 )
             prev = padded_offsets[e].item()
 
+    @skip_on_out_of_resources
     def test_round_with_fused_gather(self):
         """Token rounding + fused gather gives same result as plain fused gather."""
         from importlib import import_module
@@ -1414,6 +1446,7 @@ class TestTokenRounding:
 # =============================================================================
 
 
+@pytest.mark.slow
 class TestCombinedOptimizations:
     """Test all optimizations together."""
 
@@ -1583,6 +1616,7 @@ def _make_mock_sigmoid_moe_block(
     return moe_block, T, H, FF, E, K
 
 
+@pytest.mark.slow
 class TestHFScatterMoESigmoidRouting:
     """Test HFScatterMoEGatedMLP forward with sigmoid routing on GPU."""
 
@@ -1724,6 +1758,7 @@ class TestHFScatterMoESigmoidRouting:
         )
 
 
+@pytest.mark.slow
 class TestHFScatterMoESigmoidWithSharedExperts:
     """Test HFScatterMoEGatedMLP with sigmoid routing + shared experts."""
 

tests/e2e/integrations/test_scattermoe_lora_olmoe.py

@@ -933,7 +933,7 @@ class TestKernelizeIntegration:
     def _get_repo_path():
         """Get the path to scattermoe_lora within axolotl's plugin."""
         return (
-            Path(__file__).parent.parent.parent
+            Path(__file__).parent.parent.parent.parent
             / "src"
             / "axolotl"
             / "integrations"
@@ -1219,7 +1219,7 @@ class TestSharedExpertHandling:
 
         # Kernelize
         repo_path = (
-            Path(__file__).parent.parent.parent
+            Path(__file__).parent.parent.parent.parent
             / "src"
             / "axolotl"
             / "integrations"

tests/e2e/multigpu/solo/test_flex.py

@@ -86,5 +86,5 @@ class TestPackedFlex:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.1, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.1, "Train Loss (%s) is too high"
         )

tests/e2e/multigpu/test_dist_muon_fsdp2.py

@@ -37,7 +37,7 @@ def verify_training_success(temp_dir):
             event_file = os.path.join(tb_log_path, event_files[0])
             reader = SummaryReader(event_file)
             df = reader.scalars
-            train_loss_df = df[df.tag == "train/train_loss"]
+            train_loss_df = df[df.tag == "train/loss"]
             if len(train_loss_df) > 0:
                 final_loss = train_loss_df.value.values[-1]
                 assert not torch.isnan(torch.tensor(final_loss)), (

tests/e2e/multigpu/test_fp8_fsdp2.py

@@ -37,7 +37,7 @@ def verify_fp8_training_success(temp_dir):
             event_file = os.path.join(tb_log_path, event_files[0])
             reader = SummaryReader(event_file)
             df = reader.scalars
-            train_loss_df = df[df.tag == "train/train_loss"]
+            train_loss_df = df[df.tag == "train/loss"]
             if len(train_loss_df) > 0:
                 final_loss = train_loss_df.value.values[-1]
                 assert not torch.isnan(torch.tensor(final_loss)), (

tests/e2e/multigpu/test_fsdp1.py

@@ -38,7 +38,7 @@ def verify_training_success(temp_dir):
             event_file = os.path.join(tb_log_path, event_files[0])
             reader = SummaryReader(event_file)
             df = reader.scalars
-            train_loss_df = df[df.tag == "train/train_loss"]
+            train_loss_df = df[df.tag == "train/loss"]
             if len(train_loss_df) > 0:
                 final_loss = train_loss_df.value.values[-1]
                 assert not torch.isnan(torch.tensor(final_loss)), (

tests/e2e/multigpu/test_fsdp2.py

@@ -38,7 +38,7 @@ def verify_training_success(temp_dir):
             event_file = os.path.join(tb_log_path, event_files[0])
             reader = SummaryReader(event_file)
             df = reader.scalars
-            train_loss_df = df[df.tag == "train/train_loss"]
+            train_loss_df = df[df.tag == "train/loss"]
             if len(train_loss_df) > 0:
                 final_loss = train_loss_df.value.values[-1]
                 assert not torch.isnan(torch.tensor(final_loss)), (

tests/e2e/multigpu/test_gemma3.py

@@ -94,5 +94,5 @@ class TestMultiGPUGemma3:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.8, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 1.8, "Train Loss (%s) is too high"
         )

tests/e2e/multigpu/test_llama.py

@@ -90,7 +90,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.8, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.8, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -156,7 +156,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     def test_dpo_lora_ddp(self, temp_dir):
@@ -233,7 +233,7 @@ class TestMultiGPULlama:
         loss_threshold = 2.3
         check_tensorboard(
             temp_dir + "/runs",
-            "train/train_loss",
+            "train/loss",
             loss_threshold,
             "Train Loss (%s) is too high",
         )
@@ -312,7 +312,7 @@ class TestMultiGPULlama:
         loss_threshold = 2.3
         check_tensorboard(
             temp_dir + "/runs",
-            "train/train_loss",
+            "train/loss",
             loss_threshold,
             "Train Loss (%s) is too high",
         )
@@ -385,7 +385,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -461,7 +461,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @require_torch_2_6_0
@@ -543,7 +543,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.1, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.1, "Train Loss (%s) is too high"
         )
 
     def test_fsdp_qlora_prequant_packed(self, temp_dir):
@@ -623,7 +623,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -708,7 +708,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.45, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.45, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -784,7 +784,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -859,7 +859,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.5, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.5, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.skip(
@@ -925,5 +925,5 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 4.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 4.0, "Train Loss (%s) is too high"
         )

tests/e2e/multigpu/test_ray.py

@@ -79,7 +79,7 @@ class TestMultiGPURay:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @require_torch_2_7_0
@@ -138,7 +138,7 @@ class TestMultiGPURay:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )
 
     @require_torch_2_7_0
@@ -205,5 +205,5 @@ class TestMultiGPURay:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.3, "Train Loss (%s) is too high"
         )

tests/e2e/multigpu/test_tp.py

@@ -64,5 +64,5 @@ class TestTensorParallel:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 1.0, "Train Loss (%s) is too high"
         )

tests/e2e/patched/test_fa_xentropy.py

@@ -78,5 +78,5 @@ class TestFAXentropyLlama:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.5, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 1.5, "Train Loss (%s) is too high"
         )

tests/e2e/patched/test_flattening.py

@@ -77,5 +77,5 @@ class TestFAFlattening:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.5, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 1.5, "Train Loss (%s) is too high"
         )

tests/e2e/patched/test_lora_llama_multipack.py

@@ -4,8 +4,7 @@ E2E tests for lora llama
 
 import unittest
 
-import pytest
-from transformers.utils import is_auto_gptq_available, is_torch_bf16_gpu_available
+from transformers.utils import is_torch_bf16_gpu_available
 
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
@@ -68,51 +67,3 @@ class TestLoraLlama(unittest.TestCase):
 
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_model_output_exists(temp_dir, cfg)
-
-    @pytest.mark.skipif(not is_auto_gptq_available(), reason="auto-gptq not available")
-    @with_temp_dir
-    def test_lora_gptq_packed(self, temp_dir):
-        cfg = DictDefault(
-            {
-                "base_model": "lilmeaty/SmolLM2-135M-Instruct-GPTQ",
-                "model_type": "AutoModelForCausalLM",
-                "tokenizer_type": "AutoTokenizer",
-                "sequence_len": 1024,
-                "sample_packing": True,
-                "flash_attention": True,
-                "load_in_8bit": True,
-                "adapter": "lora",
-                "gptq": True,
-                "gptq_disable_exllama": True,
-                "lora_r": 32,
-                "lora_alpha": 64,
-                "lora_dropout": 0.05,
-                "lora_target_linear": True,
-                "val_set_size": 0.02,
-                "special_tokens": {
-                    "pad_token": "<|endoftext|>",
-                },
-                "datasets": [
-                    {
-                        "path": "mhenrichsen/alpaca_2k_test",
-                        "type": "alpaca",
-                    },
-                ],
-                "num_epochs": 2,
-                "max_steps": 20,
-                "save_steps": 0.5,
-                "micro_batch_size": 8,
-                "gradient_accumulation_steps": 1,
-                "output_dir": temp_dir,
-                "learning_rate": 0.00001,
-                "optimizer": "adamw_torch_fused",
-                "lr_scheduler": "cosine",
-                "save_first_step": False,
-            }
-        )
-        cfg = validate_config(cfg)
-        normalize_config(cfg)
-        dataset_meta = load_datasets(cfg=cfg)
-
-        train(cfg=cfg, dataset_meta=dataset_meta)
-        check_model_output_exists(temp_dir, cfg)

tests/e2e/patched/test_resume.py

@@ -9,8 +9,8 @@ import subprocess
 from transformers.utils import is_torch_bf16_gpu_available
 
 from axolotl.common.datasets import load_datasets
+from axolotl.core.trainers.constants import TOKENS_STATE_FILE
 from axolotl.train import train
-from axolotl.utils.callbacks.tokens_per_second import TOKENS_STATE_FILE
 from axolotl.utils.config import normalize_config, validate_config
 from axolotl.utils.dict import DictDefault
 

tests/e2e/patched/test_unsloth_qlora.py

@@ -73,7 +73,7 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.0, "Train Loss (%s) is too high"
         )
 
     def test_unsloth_llama_qlora_unpacked(self, temp_dir):
@@ -124,7 +124,7 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.0, "Train Loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -180,5 +180,5 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.0, "Train Loss (%s) is too high"
         )

tests/e2e/solo/test_flex.py

@@ -63,5 +63,5 @@ class TestPackedFlex(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.1, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.1, "Train Loss (%s) is too high"
         )

tests/e2e/test_deepseekv3.py

@@ -14,6 +14,9 @@ from axolotl.utils.dict import DictDefault
 from tests.hf_offline_utils import enable_hf_offline
 
 
+@pytest.mark.skip(
+    reason="DeepSeek-V3-11M remote model code needs _supports_flash_attn=True for newer transformers"
+)
 class TestDeepseekV3:
     """
     Test case for DeepseekV3 models

tests/e2e/test_dpo.py

@@ -262,6 +262,7 @@ class TestDPOLlamaLora(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_model_output_exists(Path(temp_dir) / "checkpoint-20", cfg)
 
+    @pytest.mark.skip(reason="TRL ORPO trainer has internal zip() length mismatch bug")
     @with_temp_dir
     def test_orpo_lora(self, temp_dir):
         cfg = DictDefault(

tests/e2e/test_embeddings_lr.py

@@ -57,9 +57,7 @@ class TestEmbeddingsLrScale(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_model_output_exists(temp_dir, cfg)
 
-        check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Loss is too high"
-        )
+        check_tensorboard(temp_dir + "/runs", "train/loss", 2.0, "Loss is too high")
 
     @with_temp_dir
     def test_train_w_embedding_lr(self, temp_dir):
@@ -100,6 +98,4 @@ class TestEmbeddingsLrScale(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_model_output_exists(temp_dir, cfg)
 
-        check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Loss is too high"
-        )
+        check_tensorboard(temp_dir + "/runs", "train/loss", 2.0, "Loss is too high")

tests/e2e/test_llama_pretrain.py

@@ -66,7 +66,7 @@ class TestPretrainLlama:
             loss_threshold = 6.5
         check_tensorboard(
             temp_dir + "/runs",
-            "train/train_loss",
+            "train/loss",
             loss_threshold,
             "Train Loss (%s) is too high",
         )

tests/e2e/test_mixtral.py

@@ -70,7 +70,7 @@ class TestMixtral(unittest.TestCase):
 
         model, _, _ = train(cfg=cfg, dataset_meta=dataset_meta)
         assert (
-            model.base_model.model.model.layers[0].block_sparse_moe.gate.weight.dtype
+            model.base_model.model.model.layers[0].mlp.gate.weight.dtype
             == torch.float32
         )
         check_model_output_exists(temp_dir, cfg)
@@ -125,7 +125,7 @@ class TestMixtral(unittest.TestCase):
 
         model, _, _ = train(cfg=cfg, dataset_meta=dataset_meta)
         assert (
-            model.base_model.model.model.layers[0].block_sparse_moe.gate.weight.dtype
+            model.base_model.model.model.layers[0].mlp.gate.weight.dtype
             == torch.float32
         )
         check_model_output_exists(temp_dir, cfg)
@@ -183,7 +183,7 @@ class TestMixtral(unittest.TestCase):
 
         model, _, _ = train(cfg=cfg, dataset_meta=dataset_meta)
         assert (
-            model.base_model.model.model.layers[0].block_sparse_moe.gate.weight.dtype
+            model.base_model.model.model.layers[0].mlp.gate.weight.dtype
             == torch.float32
         )
         check_model_output_exists(temp_dir, cfg)
@@ -241,7 +241,7 @@ class TestMixtral(unittest.TestCase):
 
         model, _, _ = train(cfg=cfg, dataset_meta=dataset_meta)
         assert (
-            model.base_model.model.model.layers[0].block_sparse_moe.gate.weight.dtype
+            model.base_model.model.model.layers[0].mlp.gate.weight.dtype
             == torch.float32
         )
         check_model_output_exists(temp_dir, cfg)

tests/e2e/test_optimizers.py

@@ -4,6 +4,8 @@ E2E tests for custom optimizers using Llama
 
 import unittest
 
+import pytest
+
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
 from axolotl.utils.config import normalize_config, validate_config
@@ -282,3 +284,60 @@ class TestCustomOptimizers(unittest.TestCase):
 
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_model_output_exists(temp_dir, cfg)
+
+
+@require_torch_2_7_0
+@pytest.mark.parametrize(
+    "optimizer_name,expected_class,learning_rate",
+    [
+        ("flash_adamw", "FlashAdamW", 0.00001),
+        ("flash_adam", "FlashAdam", 0.00001),
+        ("flash_sgd", "FlashSGD", 0.01),
+        ("flash_sgdw", "FlashSGDW", 0.01),
+        ("flash_lion", "FlashLion", 0.0001),
+    ],
+)
+def test_flash_optimizers(tmp_path, optimizer_name, expected_class, learning_rate):
+    pytest.importorskip("flashoptim")
+    temp_dir = str(tmp_path)
+    cfg = DictDefault(
+        {
+            "base_model": "HuggingFaceTB/SmolLM2-135M",
+            "model_type": "AutoModelForCausalLM",
+            "tokenizer_type": "AutoTokenizer",
+            "sequence_len": 1024,
+            "load_in_8bit": True,
+            "adapter": "lora",
+            "lora_r": 8,
+            "lora_alpha": 16,
+            "lora_dropout": 0.05,
+            "lora_target_linear": True,
+            "val_set_size": 0.02,
+            "special_tokens": {
+                "pad_token": "<|endoftext|>",
+            },
+            "datasets": [
+                {
+                    "path": "mhenrichsen/alpaca_2k_test",
+                    "type": "alpaca",
+                },
+            ],
+            "num_epochs": 1,
+            "micro_batch_size": 8,
+            "gradient_accumulation_steps": 1,
+            "output_dir": temp_dir,
+            "learning_rate": learning_rate,
+            "optimizer": optimizer_name,
+            "max_steps": 5,
+            "lr_scheduler": "cosine",
+            "save_first_step": False,
+        }
+    )
+
+    cfg = validate_config(cfg)
+    normalize_config(cfg)
+    dataset_meta = load_datasets(cfg=cfg)
+
+    _, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
+    check_model_output_exists(temp_dir, cfg)
+    assert trainer.optimizer.optimizer.__class__.__name__ == expected_class

tests/e2e/test_packing_loss.py

@@ -62,5 +62,5 @@ class TestPackedLlama(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss (%s) is too high"
+            temp_dir + "/runs", "train/loss", 2.0, "Train Loss (%s) is too high"
         )