code review feedback

* Train parameters exclusively in specific ranges

* Fix the style and update docs

* Update yaml example

README.md

@@ -25,7 +25,7 @@ Features:
 - [Environment](#environment)
   - [Docker](#docker)
   - [Conda/Pip venv](#condapip-venv)
-  - [Cloud GPU](#cloud-gpu) - Latitude.sh, RunPod
+  - [Cloud GPU](#cloud-gpu) - Latitude.sh, JarvisLabs, RunPod
   - [Bare Metal Cloud GPU](#bare-metal-cloud-gpu)
   - [Windows](#windows)
   - [Launching on public clouds via SkyPilot](#launching-on-public-clouds-via-skypilot)
@@ -199,6 +199,7 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --
 For cloud GPU providers that support docker images, use [`winglian/axolotl-cloud:main-latest`](https://hub.docker.com/r/winglian/axolotl-cloud/tags)
 
 - on Latitude.sh use this [direct link](https://latitude.sh/blueprint/989e0e79-3bf6-41ea-a46b-1f246e309d5c)
+- on JarvisLabs.ai use this [direct link](https://jarvislabs.ai/templates/axolotl)
 - on RunPod use this [direct link](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
 
 #### Bare Metal Cloud GPU
@@ -1079,6 +1080,10 @@ fsdp_config:
   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```
 
+##### FSDP + QLoRA
+
+Axolotl supports training with FSDP and QLoRA, see [these docs](docs/fsdp_qlora.md) for more information.
+
 ##### Weights & Biases Logging
 
 Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.
@@ -1298,4 +1303,6 @@ consider sponsoring the project via [GitHub Sponsors](https://github.com/sponsor
 
 #### 🥉 Bronze Sponsors - $500/mo
 
+ - [JarvisLabs.ai](https://jarvislabs.ai)
+
 ---

deepspeed_configs/zero1.json

@@ -16,6 +16,7 @@
     "min_loss_scale": 1
   },
   "gradient_accumulation_steps": "auto",
+  "gradient_clipping": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

deepspeed_configs/zero2.json

@@ -20,6 +20,7 @@
     "min_loss_scale": 1
   },
   "gradient_accumulation_steps": "auto",
+  "gradient_clipping": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

deepspeed_configs/zero3.json

@@ -24,6 +24,7 @@
     "min_loss_scale": 1
   },
   "gradient_accumulation_steps": "auto",
+  "gradient_clipping": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

deepspeed_configs/zero3_bf16.json

@@ -24,6 +24,7 @@
     "min_loss_scale": 1
   },
   "gradient_accumulation_steps": "auto",
+  "gradient_clipping": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

docs/fsdp_qlora.md

@@ -0,0 +1,37 @@
+# FDSP + QLoRA
+
+## Background
+
+Using FSDP with QLoRA is essential for **fine-tuning larger (70b+ parameter) LLMs on consumer GPUs.**  For example, you can use FSDP + QLoRA to train a 70b model on two 24GB GPUs[^1].
+
+Below, we describe how to use this feature in Axolotl.
+
+## Usage
+
+To enable `QLoRA` with `FSDP`, you need to perform the following steps:
+
+> ![Tip]
+> See the [example config](#example-config) file in addition to reading these instructions.
+
+1. Set `adapter: qlora` in your axolotl config file.
+2. Enable FSDP in your axolotl config, as [described here](https://github.com/OpenAccess-AI-Collective/axolotl?tab=readme-ov-file#fsdp).
+3. Use one of the supported model types: `llama`, `mistral` or `mixtral`.
+
+## Example Config
+
+[examples/llama-2/qlora-fsdp.yml](../examples/llama-2/qlora-fsdp.yml) contains an example of how to enable QLoRA + FSDP in axolotl.
+
+## References
+
+- [PR #1378](https://github.com/OpenAccess-AI-Collective/axolotl/pull/1378) enabling QLoRA in FSDP in Axolotl.
+- [Blog Post](https://www.answer.ai/posts/2024-03-06-fsdp-qlora.html) from the [Answer.AI](https://www.answer.ai/) team describing the work that enabled QLoRA in FSDP.
+- Related HuggingFace PRs Enabling FDSP + QLoRA:
+    - Accelerate [PR#2544](https://github.com/huggingface/accelerate/pull/2544 )
+    - Transformers [PR#29587](https://github.com/huggingface/transformers/pull/29587)
+    - TRL [PR#1416](https://github.com/huggingface/trl/pull/1416)
+    - PEFT [PR#1550](https://github.com/huggingface/peft/pull/1550)
+
+
+
+
+[^1]: This was enabled by [this work](https://www.answer.ai/posts/2024-03-06-fsdp-qlora.html) from the Answer.AI team.

examples/gemma/qlora.yml

@@ -21,7 +21,7 @@ lora_dropout: 0.05
 lora_target_linear: true
 
 sequence_len: 4096
-sample_packing: true
+sample_packing: false
 pad_to_sequence_len: true
 
 wandb_project:

examples/llama-2/qlora-fsdp.yml

@@ -0,0 +1,70 @@
+base_model: NousResearch/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: yahma/alpaca-cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 512
+sample_packing: false
+pad_to_sequence_len: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 4
+num_epochs: 4
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 0.00001
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+fp16:
+tf32: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch: 4
+eval_table_size:
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+  - full_shard
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
+special_tokens:

examples/mistral/mixtral-qlora-fsdp.yml

@@ -0,0 +1,74 @@
+base_model: mistralai/Mixtral-8x7B-v0.1
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+output_dir: ./qlora-out
+
+model_config:
+  output_router_logits: true
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 1024
+sample_packing: false
+pad_to_sequence_len: false
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+fp16:
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_steps: 10
+evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
+saves_per_epoch: 1
+debug:
+weight_decay: 0.0
+fsdp:
+  - full_shard
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: MixtralSparseMoeBlock
+special_tokens:

examples/mistral/mixtral.yml

@@ -16,12 +16,12 @@ output_dir: ./qlora-out
 
 ## You can optionally freeze the entire model and unfreeze a subset of parameters
 unfrozen_parameters:
-#  - lm_head.*
-#  - model.embed_tokens.*
-#  - model.layers.2[0-9]+.block_sparse_moe.gate.*
-#  - model.layers.2[0-9]+.block_sparse_moe.experts.*
-#  - model.layers.3[0-9]+.block_sparse_moe.gate.*
-#  - model.layers.3[0-9]+.block_sparse_moe.experts.*
+#  - ^lm_head.weight$
+#  - ^model.embed_tokens.weight$[:32000]
+#  - model.layers.2[0-9]+.block_sparse_moe.gate
+#  - model.layers.2[0-9]+.block_sparse_moe.experts
+#  - model.layers.3[0-9]+.block_sparse_moe.gate
+#  - model.layers.3[0-9]+.block_sparse_moe.experts
 
 model_config:
   output_router_logits: true

requirements.txt

@@ -3,7 +3,7 @@ packaging==23.2
 peft==0.9.0
 transformers==4.38.2
 tokenizers==0.15.0
-bitsandbytes>=0.41.1
+bitsandbytes>=0.43.0
 accelerate==0.26.1
 deepspeed==0.13.1
 pydantic==2.6.3
@@ -40,3 +40,4 @@ gcsfs
 # adlfs
 
 trl>=0.7.9
+fastcore>=1.5.29

src/axolotl/core/policies/auto_wrap.py

@@ -0,0 +1,55 @@
+"""module for building the auto wrap policy for FSDP"""
+import functools
+
+from peft import PrefixEncoder, PromptEmbedding, PromptEncoder
+from torch.distributed.fsdp.wrap import (
+    _or_policy,
+    lambda_auto_wrap_policy,
+    transformer_auto_wrap_policy,
+)
+from transformers.models.llama.modeling_llama import LlamaDecoderLayer
+from transformers.models.mistral.modeling_mistral import MistralDecoderLayer
+from transformers.models.mixtral.modeling_mixtral import MixtralDecoderLayer
+
+SUPPORTED_AUTO_WRAP_MODEL_TYPES = [
+    "llama",
+    "mistral",
+    "mixtral",
+]
+
+
+def get_wrapping_policy_factory(model_type):
+    if model_type == "llama":
+        layer_to_wrap = LlamaDecoderLayer
+    elif model_type == "mistral":
+        layer_to_wrap = MistralDecoderLayer
+    elif model_type == "mixtral":
+        layer_to_wrap = MixtralDecoderLayer
+
+    def get_wrapping_policy():
+        """This checks for lora layers (has weight and requires_grad)"""
+
+        def lambda_policy_fn(module):
+            return (
+                len(list(module.named_children())) == 0
+                and getattr(module, "weight", None) is not None
+                and module.weight.requires_grad
+            )
+
+        lambda_policy = functools.partial(
+            lambda_auto_wrap_policy, lambda_fn=lambda_policy_fn
+        )
+        transformer_layer_name = layer_to_wrap
+        transformer_wrap_policy = functools.partial(
+            transformer_auto_wrap_policy,
+            transformer_layer_cls=(
+                PrefixEncoder,
+                PromptEncoder,
+                PromptEmbedding,
+                transformer_layer_name,
+            ),
+        )
+        policies = [lambda_policy, transformer_wrap_policy]
+        return functools.partial(_or_policy, policies=policies)
+
+    return get_wrapping_policy

src/axolotl/core/trainer_builder.py

@@ -8,6 +8,7 @@ import importlib
 import importlib.util
 import logging
 import math
+import os
 import sys
 from abc import abstractmethod
 from dataclasses import dataclass, field
@@ -17,7 +18,10 @@ from typing import List, Optional, Type, Union
 
 import torch
 import transformers
+from accelerate import FullyShardedDataParallelPlugin
+from accelerate.utils import str_to_bool
 from datasets import Dataset
+from torch.distributed.fsdp import MixedPrecision
 from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
 from transformers import (
@@ -30,6 +34,7 @@ from transformers.trainer_utils import seed_worker
 from transformers.utils import is_sagemaker_mp_enabled
 from trl import DPOTrainer
 
+from axolotl.core.policies.auto_wrap import get_wrapping_policy_factory
 from axolotl.loraplus import create_loraplus_optimizer
 from axolotl.monkeypatch.multipack import SUPPORTED_MULTIPACK_MODEL_TYPES
 from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
@@ -191,6 +196,10 @@ class AxolotlTrainingArguments(TrainingArguments):
         default=1e-6,
         metadata={"help": "loraplus learning rate for lora embedding layers."},
     )
+    qlora: bool = field(
+        default=False,
+        metadata={"help": "whether this is a qlora training"},
+    )
 
 
 class AxolotlTrainer(Trainer):
@@ -468,6 +477,56 @@ class AxolotlTrainer(Trainer):
 
         return super().push_to_hub(*args, **kwargs)
 
+    @wraps(Trainer.create_accelerator_and_postprocess)
+    def create_accelerator_and_postprocess(self):
+        rank = int(os.environ.get("LOCAL_RANK", 0))
+        res = super().create_accelerator_and_postprocess()
+
+        if self.args.qlora is False:
+            return res
+
+        # the rest of this method override is specific to fsdp + qlora (for now)
+        sync_module_states = (
+            str_to_bool(os.environ.get("FSDP_SYNC_MODULE_STATES", "True")) == 1
+        )
+
+        mp_policy = None
+        amp = os.environ["ACCELERATE_MIXED_PRECISION"]
+        if amp == "fp16":
+            mp_policy = MixedPrecision(
+                param_dtype=torch.float32,
+                reduce_dtype=torch.float32,
+                buffer_dtype=torch.float32,
+            )
+        elif amp == "bf16":
+            mp_policy = MixedPrecision(
+                param_dtype=torch.float32,
+                reduce_dtype=torch.float32,
+                buffer_dtype=torch.float32,
+            )
+
+        # If somehow we figure out how we want to parameterize we want to autocast buffers...
+        # mp_policy = MixedPrecision(param_dtype=torch.bfloat16, reduce_dtype=torch.bfloat16, buffer_dtype=torch.float32)
+        # load_param_skip_names = ['inv_freq']
+
+        if self.is_fsdp_enabled:
+            wrapping_policy = get_wrapping_policy_factory(self.args.model_type)
+            fsdp_plugin = FullyShardedDataParallelPlugin(
+                auto_wrap_policy=wrapping_policy(),
+                cpu_offload=False,
+                use_orig_params=False,
+                limit_all_gathers=True,
+                param_init_fn=lambda module: module.to_empty(
+                    device=torch.device("cuda"), recurse=False
+                )
+                if (rank != 0 and sync_module_states)
+                else None,
+                mixed_precision_policy=mp_policy,
+            )
+            self.accelerator.state.fsdp_plugin = fsdp_plugin
+
+        return res
+
 
 class AxolotlMambaTrainer(AxolotlTrainer):
     """
@@ -787,6 +846,9 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
             if self.cfg.fsdp_config:
                 training_arguments_kwargs["fsdp_config"] = dict(self.cfg.fsdp_config)
 
+        if self.cfg.adapter == "qlora":
+            training_arguments_kwargs["qlora"] = True
+
         # deepspeed
         if self.cfg.deepspeed:
             training_arguments_kwargs["deepspeed"] = self.cfg.deepspeed

src/axolotl/logging_config.py

@@ -30,6 +30,7 @@ class ColorfulFormatter(Formatter):
 
 DEFAULT_LOGGING_CONFIG: Dict[str, Any] = {
     "version": 1,
+    "disable_existing_loggers": False,
     "formatters": {
         "simple": {
             "format": "[%(asctime)s] [%(levelname)s] [%(name)s.%(funcName)s:%(lineno)d] [PID:%(process)d] %(message)s",

src/axolotl/monkeypatch/moe/linear.py

@@ -0,0 +1,147 @@
+"""
+Adapted from:
+https://github.com/shawntan/scattermoe
+https://arxiv.org/abs/2403.08245
+"""
+
+import torch
+import torch.nn as nn
+from axolotl.monkeypatch.moe import ops
+
+class ParallelLinear(torch.autograd.Function):
+    @staticmethod
+    def forward(
+        ctx, x, expert_weights, k,
+        sorted_expert_idxs, sorted_scattered_idxs,
+        padded_block_idxs, expert_offsets,
+        gates=None, grouped_in=False, grouped_out=False,
+    ):
+
+        output = ops.scatter2scatter(
+            X=x, W=expert_weights,
+            sorted_expert_idxs=sorted_expert_idxs,
+            sorted_scattered_idxs=sorted_scattered_idxs,
+            padded_block_idxs=padded_block_idxs,
+            k=k, x_grouped=grouped_in, y_grouped=grouped_out
+        )
+        if gates is not None:
+            output_expanded = output.view(gates.size(0), gates.size(1), output.size(-1))
+            output = torch.bmm(
+                gates[:, None, :],
+                output_expanded
+            ).squeeze(1)
+        else:
+            output_expanded = None
+
+        ctx.save_for_backward(
+            x, expert_weights,
+            sorted_expert_idxs,
+            sorted_scattered_idxs,
+            padded_block_idxs, expert_offsets,
+            gates,
+            output_expanded
+        )
+        ctx.grouped_in = grouped_in
+        ctx.grouped_out = grouped_out
+        ctx.k = k
+        return output
+    @staticmethod
+    def backward(ctx, grad_out):
+        (x, expert_weights,
+         sorted_expert_idxs,
+         sorted_scattered_idxs,
+         padded_block_idxs, expert_offsets,
+         gates, output_expanded) = ctx.saved_tensors
+        k = ctx.k
+        grouped_in = ctx.grouped_in
+        grouped_out = ctx.grouped_out
+        # print("backward")
+        if gates is not None:
+            # calculate gates gradient
+            d_gates = torch.bmm(output_expanded, grad_out[:, :, None]).squeeze(-1)
+            gates_flat = gates.flatten()
+            gate_fan = gates.size(1)
+            # print("expanded and grouping")
+            grouped_grad_out = output_expanded.flatten(0, 1) # reuse expanded buffer later
+        else:
+            d_gates = None
+            gates_flat = None
+            gate_fan = 1
+            grouped_grad_out = None
+
+        if grouped_out:
+            grouped_grad_out = grad_out
+        else:
+            grouped_grad_out = ops.group(grad_out, sorted_scattered_idxs,
+                                                 fan_out=gate_fan, coeff=gates_flat,
+                                                 out=grouped_grad_out)
+        if grouped_in:
+            grouped_x = x
+            d_expanded_input = None
+        else:
+            grouped_x = ops.group(x, sorted_scattered_idxs, fan_out=k)
+            d_expanded_input = grouped_x
+        d_weights = ops.group_bwd_W(
+            DY=grouped_grad_out, X=grouped_x,
+            expert_offsets=expert_offsets,
+            E=expert_weights.size(0)
+        )
+        d_expanded_input = ops.scatter2scatter(
+            X=grouped_grad_out, x_grouped=True,
+            W=expert_weights.permute(0, 2, 1),
+            padded_block_idxs=padded_block_idxs,
+            sorted_expert_idxs=sorted_expert_idxs,
+            sorted_scattered_idxs=sorted_scattered_idxs,
+            k=1,
+            y_grouped=grouped_in,
+            out=d_expanded_input # Reuse grouped_x buffer
+        )
+
+        if k == 1:
+            d_input = d_expanded_input
+        else:
+            d_input = d_expanded_input.view(x.size(0), k, d_expanded_input.size(-1)).sum(-2)
+        # print("backward end.")
+        return (
+            # x, expert_weights, k,
+            d_input, d_weights, None,
+            # sorted_expert_idxs, sorted_scattered_idxs,
+            None, None,
+            # padded_block_idxs, expert_offsets,
+            None, None,
+            # gates
+            d_gates, None, None
+        )
+
+def parallel_linear(inputs, expert_weights, k,
+                    sorted_expert_idxs, sorted_scattered_idxs,
+                    padded_block_idxs, expert_offsets,
+                    gates=None):
+    results = ParallelLinear.apply(inputs, expert_weights, k,
+                                   sorted_expert_idxs, sorted_scattered_idxs,
+                                   padded_block_idxs, expert_offsets, gates)
+    return results
+
+class ParallelExperts(nn.Module):
+    def __init__(self, num_experts, input_size, output_size) -> None:
+        super().__init__()
+        self.weight = nn.Parameter(torch.empty(num_experts, output_size, input_size))
+        self.num_experts = num_experts
+        self.input_size = input_size
+        self.output_size = output_size
+
+    def extra_repr(self):
+        return 'num_experts={}, input_size={}, output_size={}'.format(
+            self.num_experts, self.input_size, self.output_size)
+
+    def forward(self, inputs, k, sorted_expert_idxs, sorted_scattered_idxs,
+                padded_block_idxs, expert_offsets,
+                gates=None, grouped_in=False, grouped_out=False):
+
+        results = ParallelLinear.apply(
+            inputs, self.weight.permute(0, 2, 1), k,
+            sorted_expert_idxs, sorted_scattered_idxs,
+            padded_block_idxs, expert_offsets,
+            gates, grouped_in, grouped_out
+        )
+        return results

src/axolotl/monkeypatch/moe/mlp.py

@@ -0,0 +1,90 @@
+"""
+Adapted from:
+https://github.com/shawntan/scattermoe
+https://arxiv.org/abs/2403.08245
+"""
+
+import torch
+from torch import nn
+
+from axolotl.monkeypatch.moe import ops
+from axolotl.monkeypatch.moe.linear import ParallelExperts
+
+
+class FusedExperts(nn.Module):
+    def __init__(
+        self,
+        experts=None,
+        hidden_dim=128,
+        ffn_dim=512,
+        num_experts=8,
+        top_k=2,
+        activation=nn.SiLU(),
+    ):
+        """
+        This implements fused experts that are compatible with Mixtral.
+        MLP of type Gated-Linear Unit, typically with a SiLU activation function.
+        """
+        super(FusedExperts, self).__init__()
+
+        self.num_experts = num_experts
+        self.hidden_dim = hidden_dim
+        self.ffn_dim = ffn_dim
+        self.experts = ParallelExperts(num_experts, hidden_dim, 2 * ffn_dim)
+        self.output_experts = ParallelExperts(num_experts, ffn_dim, hidden_dim)
+        self.top_k = min(top_k, self.num_experts)
+        self.activation = activation
+
+        # parallelize all w1 and w3 computation by concat + stack
+        with torch.no_grad():
+            torch.stack(
+                [
+                    torch.cat([experts[i].w1.weight, experts[i].w3.weight], dim=0)
+                    for i in range(len(experts))
+                ],
+                dim=0,
+                out=self.experts.weight.data,
+            )
+
+            # parallelize all w2 computation by stack
+            torch.stack(
+                [expert.w2.weight for expert in experts],
+                dim=0,
+                out=self.output_experts.weight.data,
+            )
+
+    def forward(
+        self, x: torch.Tensor, routing_weights: torch.Tensor, selected_experts: torch.Tensor
+    ):
+        x_shape = x.size()
+        x = x.view(-1, x_shape[-1])
+        with torch.no_grad():
+            sorted_expert_idxs, sorted_scattered_idxs = ops.flatten_and_sort(
+                selected_experts
+            )
+            padded_block_idxs, expert_offsets = ops.padded_block_indices(
+                sorted_expert_idxs, self.num_experts
+            )
+
+        h, gates = self.experts(
+            x,
+            self.top_k,
+            sorted_expert_idxs,
+            sorted_scattered_idxs,
+            padded_block_idxs,
+            expert_offsets,
+            grouped_out=True,
+        ).chunk(2, dim=-1)
+        h = self.activation(gates) * h
+        y = self.output_experts(
+            h,
+            1,
+            sorted_expert_idxs,
+            sorted_scattered_idxs,
+            padded_block_idxs,
+            expert_offsets,
+            grouped_in=True,
+            gates=routing_weights,
+        )
+        y = y.view(*x_shape[:-1], y.size(-1))
+        return y

src/axolotl/monkeypatch/moe/moe.py

@@ -0,0 +1,73 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from axolotl.monkeypatch.moe.mlp import FusedExperts
+
+class SparseMoeBlock(nn.Module):
+    def __init__(self, experts, gate, hidden_dim, ffn_dim, num_experts, top_k):
+        super().__init__()
+        self.hidden_dim = hidden_dim
+        self.ffn_dim = ffn_dim
+        self.num_experts = num_experts
+        self.top_k = top_k
+        self.gate = gate
+        self.experts = FusedExperts(
+            experts=experts,
+            hidden_dim=hidden_dim,
+            ffn_dim=ffn_dim,
+            num_experts=num_experts,
+            top_k=top_k,
+            activation=experts[0].act_fn
+        )
+
+    def _post_training(self, model, name):
+        # Get original weights back: reverse the concat + stack in the fused experts
+        w1s, w3s = torch.split(torch.unbind(self.experts.experts.weight, dim=0), 2, dim=1)
+        w2s = torch.unbind(self.experts.output_experts.weight, dim=0)
+
+        # Recreate the structure of the original MixtralSparseMoeBlock
+        original_moe = nn.Module()
+        original_moe.hidden_dim = self.hidden_dim
+        original_moe.ffn_dim = self.ffn_dim
+        original_moe.num_experts = self.num_experts
+        original_moe.top_k = self.top_k
+
+        # Recreate the gating module
+        original_moe.gate = nn.Linear(self.hidden_dim, self.num_experts, bias=False)
+        original_moe.gate.weight.data = self.gate.weight.data
+
+        # Recreate the experts as a ModuleList
+        original_moe.experts = nn.ModuleList()
+        for expert_idx in range(self.num_experts):
+            expert = nn.Module()
+            expert.w1 = nn.Linear(self.hidden_dim, 2 * self.ffn_dim, bias=False)
+            expert.w2 = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)
+            expert.w3 = nn.Linear(self.hidden_dim, 2 * self.ffn_dim, bias=False)
+            expert.act_fn = self.experts.activation
+
+            expert.w1.weight.data = torch.cat([w1s[expert_idx], w3s[expert_idx]], dim=0)
+            expert.w2.weight.data = w2s[expert_idx]
+
+            original_moe.experts.append(expert)
+
+        # Replace the SparseMoeBlock with the recreated MixtralSparseMoeBlock structure
+        setattr(model, name, original_moe)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = self.gate(hidden_states)
+        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
+        routing_weights, selected_experts = torch.topk(routing_weights, self.top_k, dim=-1)
+        routing_weights /= routing_weights.sum(dim=-1, keepdim=True)
+
+        # we cast back to the input dtype
+        routing_weights = routing_weights.to(hidden_states.dtype)
+
+        # Fused expert forward
+        final_hidden_states = self.experts(hidden_states, routing_weights, selected_experts)
+
+        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+        return final_hidden_states, router_logits

src/axolotl/monkeypatch/moe/ops.py

@@ -0,0 +1,353 @@
+"""
+Adapted from:
+https://github.com/shawntan/scattermoe
+https://arxiv.org/abs/2403.08245
+"""
+
+import torch
+import triton
+import triton.language as tl
+from torch.nn import functional as F
+
+BLOCK_M = 128
+
+@torch.jit.script
+def flatten_and_sort(expert_idxs:torch.Tensor):
+    flattened_expert_idxs = expert_idxs.flatten()
+    sorted_expert_idxs, sorted_scattered_idxs = torch.sort(flattened_expert_idxs)
+    return sorted_expert_idxs, sorted_scattered_idxs
+
+@torch.jit.script
+def padded_block_indices(sorted_experts_idxs: torch.Tensor, k: int, N_BLOCK_SIZE: int=BLOCK_M) :
+    expert_counts = torch.bincount(sorted_experts_idxs, minlength=k)
+    padded_block_counts = ((expert_counts - 1) // N_BLOCK_SIZE) + 1
+    padded_expert_block_end = padded_block_counts.cumsum(-1)
+    expert_boundaries_end = expert_counts.cumsum(-1)
+    expert_boundaries_start = expert_boundaries_end - expert_counts
+    padded_expert_block_start = padded_expert_block_end - padded_block_counts
+    block_idxs = torch.arange(padded_expert_block_end[-1],
+                              dtype=sorted_experts_idxs.dtype,
+                              device=sorted_experts_idxs.device)
+    block_mask = (
+        (block_idxs[:, None] < padded_expert_block_start) |
+        (block_idxs[:, None] >= padded_expert_block_end)
+    )
+    expanded_block_idxs = (
+        N_BLOCK_SIZE * (block_idxs[:, None] - padded_expert_block_start) +
+        expert_boundaries_start
+    )
+    expanded_block_idxs = expanded_block_idxs.masked_fill(block_mask, 0).sum(-1)
+    return expanded_block_idxs, expert_boundaries_end
+
+
+
+def _scatter2scatter_configs():
+    return [
+        triton.Config({'BLOCK_N': 128, 'BLOCK_K': 32}, num_stages=4, num_warps=4),
+    ]
+
+@triton.autotune(configs=_scatter2scatter_configs(), key=['M', 'N', 'K'], )
+@triton.heuristics({
+    "NO_K_MASK": lambda args: (args['K'] % args['BLOCK_K']) == 0,
+    "NO_N_MASK": lambda args: (args['N'] % args['BLOCK_N']) == 0,
+})
+@triton.jit
+def _scatter2scatter(
+    X_ptr, stride_xm, stride_xk,
+    W_ptr, stride_we, stride_wk, stride_wn,
+    Y_ptr, stride_ym, stride_yn,
+    grouped_idx_ptr, expert_idxs_ptr, block_start_idx_ptr,
+    FAN_OUT: tl.constexpr,
+    M: tl.constexpr, K: tl.constexpr, N: tl.constexpr, E: tl.constexpr,
+    BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
+    ACC_TYPE: tl.constexpr,
+    OUT_M: tl.constexpr,
+    allow_tf32: tl.constexpr,
+    x_grouped: tl.constexpr, y_grouped: tl.constexpr,
+    NO_K_MASK: tl.constexpr, NO_N_MASK: tl.constexpr
+):
+    pid = tl.program_id(axis=0)
+
+    N_BLOCK_COUNT = tl.cdiv(N, BLOCK_N)
+    M_block_id = pid // N_BLOCK_COUNT
+    N_block_id = pid % N_BLOCK_COUNT
+    M_range = tl.arange(0, BLOCK_M)
+    block_start_idx = tl.load(block_start_idx_ptr + M_block_id)
+    # M_block = tl.max_contiguous((block_start_idx + M_range) % OUT_M, BLOCK_M)
+    M_block = tl.max_contiguous(block_start_idx + M_range, BLOCK_M)
+    E_idxs = tl.load(expert_idxs_ptr + M_block, mask=M_block < (FAN_OUT * M), other=E)
+    E_idx = tl.min(E_idxs)
+    E_mask = E_idxs == E_idx
+    M_idx = tl.load(grouped_idx_ptr + M_block, mask=E_mask, other=0)
+    if x_grouped:
+        M_in_idx = M_block
+    else:
+        M_in_idx = M_idx // FAN_OUT
+
+    if y_grouped:
+        M_out_idx = M_block
+    else:
+        M_out_idx = M_idx
+
+    K_block = tl.arange(0, BLOCK_K)
+
+    N_block = N_block_id * BLOCK_N  + tl.arange(0, BLOCK_N)
+    N_mask = N_block < N
+    # N_block = tl.max_contiguous(tl.multiple_of(N_block % N, BLOCK_N), BLOCK_N)
+    # N_block = N_block_id * BLOCK_N + tl.arange(0, BLOCK_N)
+
+    X_blk_ptrs = X_ptr + M_in_idx[:, None] * stride_xm + K_block[None, :] * stride_xk
+    W_blk_ptrs = W_ptr + K_block[:, None] * stride_wk + N_block[None, :] * stride_wn + E_idx * stride_we
+
+    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=ACC_TYPE)
+    iters = tl.cdiv(K, BLOCK_K)
+    for K_block_id in range(0, iters):
+        if NO_K_MASK:
+            x = tl.load(X_blk_ptrs, mask=E_mask[:, None])
+            if NO_N_MASK:
+                w = tl.load(W_blk_ptrs)
+            else:
+                w = tl.load(W_blk_ptrs, mask=N_mask[None, :])
+        else:
+            K_mask = (K_block_id * BLOCK_K + K_block) < K
+            x = tl.load(X_blk_ptrs, mask=E_mask[:, None] & K_mask[None, :])
+            w = tl.load(W_blk_ptrs, mask=K_mask[:, None] & N_mask[None, :])
+        X_blk_ptrs += BLOCK_K * stride_xk
+        W_blk_ptrs += BLOCK_K * stride_wk
+        acc += tl.dot(x, w, allow_tf32=allow_tf32, out_dtype=ACC_TYPE)
+
+    Y_blk_ptrs = Y_ptr + (M_out_idx[:, None] * stride_ym + N_block[None, :] * stride_yn)
+    tl.store(Y_blk_ptrs, acc, mask=E_mask[:, None] & N_mask[None, :])
+
+def scatter2scatter(X, W, sorted_expert_idxs, sorted_scattered_idxs, k,
+                    padded_block_idxs, x_grouped=False, y_grouped=False,
+                    out=None):
+    assert sorted_scattered_idxs.size(0) == sorted_expert_idxs.size(0)
+    assert sorted_scattered_idxs.size(0) == X.size(0) * k
+    # Pre-kernel setup
+    x_dim = X.size(-1)
+    y_dim = W.size(-1)
+    L_scattered = sorted_expert_idxs.size(0)
+    if out is None:
+        O = torch.empty((L_scattered, y_dim), device=X.device, dtype=X.dtype)
+    else:
+        assert out.size(0) == L_scattered and out.size(1) == y_dim
+        O = out
+
+    def grid(META):
+        grid_num = (
+            padded_block_idxs.size(0) *
+            triton.cdiv(META['N'], META['BLOCK_N']),
+        )
+        return grid_num
+    """
+    print("X", X.size(), X.stride(),
+          "W", W.size(), W.stride(),
+          "O", O.size(), O.stride(),
+          "sorted_idxs", sorted_scattered_idxs.size(),
+          "FAN_OUT", k,
+          "BLOCK_M", BLOCK_M,
+          "grouped", (x_grouped, y_grouped))
+    """
+    _scatter2scatter[grid](
+        # X_ptr, stride_xm, stride_xk,
+        X, X.stride(0), X.stride(1),
+        # W_ptr, stride_we, stride_wk, stride_wn,
+        W, W.stride(0), W.stride(1), W.stride(2),
+        # Y_ptr, stride_ym, stride_yn,
+        O, O.stride(0), O.stride(1),
+        grouped_idx_ptr=sorted_scattered_idxs,
+        expert_idxs_ptr=sorted_expert_idxs,
+        block_start_idx_ptr=padded_block_idxs,
+        FAN_OUT=k,
+        M=X.size(0),
+        K=X.size(1),
+        N=O.size(1), E=W.size(0),
+        BLOCK_M=BLOCK_M,
+        ACC_TYPE=tl.float32,
+        OUT_M=O.size(0),
+        allow_tf32=True,
+        x_grouped=x_grouped, y_grouped=y_grouped,
+    )
+    return O
+
+
+def _config_XtY():
+    return [
+        triton.Config({'BLOCK_N': 128, 'BLOCK_K': 128, 'BLOCK_M': 32}, num_stages=4, num_warps=4),
+    ]
+
+def group_bwd_W(DY, X, expert_offsets, E):
+    DWt = torch.zeros((E, DY.size(-1), X.size(-1)), device=DY.device, dtype=DY.dtype)
+    DW = DWt.permute(0, 2, 1)
+    def grid(META):
+        grid = (
+            E * triton.cdiv(META['K'], META['BLOCK_K']),
+            triton.cdiv(META['N'], META['BLOCK_N']),
+        )
+        return grid
+    _groupXtY[grid](
+        # DY_ptr, stride_dym, stride_dyk,
+        DY, DY.stride(0), DY.stride(1),
+        # X_ptr, stride_xm, stride_xn,
+        X, X.stride(0), X.stride(1),
+        # DW_ptr, stride_dwe, stride_dwk, stride_dwn,
+        DW, DW.stride(0), DW.stride(1), DW.stride(2),
+        # expert_offsets_ptr,
+        expert_offsets,
+        # K: tl.constexpr, N: tl.constexpr,
+        M=DY.size(0), N=DY.size(-1), K=X.size(-1),
+        # ACC_TYPE: tl.constexpr,
+        ACC_TYPE=tl.float32,
+        allow_tf32=True
+    )
+    return DW
+
+@triton.autotune(configs=_config_XtY(), key=['M', 'N', 'K'], )
+@triton.heuristics({
+    "NO_K_MASK": lambda args: (args['K'] % args['BLOCK_K']) == 0,
+    "NO_N_MASK": lambda args: (args['N'] % args['BLOCK_N']) == 0,
+})
+@triton.jit
+def _groupXtY(
+    DY_ptr, stride_dym, stride_dyk,
+    X_ptr, stride_xm, stride_xn,
+    DW_ptr, stride_dwe, stride_dwk, stride_dwn,
+    expert_offsets_ptr,
+    M: tl.constexpr, K: tl.constexpr, N: tl.constexpr,
+    BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
+    ACC_TYPE: tl.constexpr,
+    allow_tf32: tl.constexpr,
+    NO_K_MASK: tl.constexpr, NO_N_MASK: tl.constexpr
+):
+    pid0 = tl.program_id(axis=0)
+    pid1 = tl.program_id(axis=1)
+    num0 = tl.num_programs(0)
+    num1 = tl.num_programs(1)
+    pid1, pid0 = tl.swizzle2d(pid1, pid0, num1, num0, 128)
+
+    K_BLOCK_COUNT = tl.cdiv(K, BLOCK_K)
+    E_idx = pid0 // K_BLOCK_COUNT
+    K_block_id = pid0 % K_BLOCK_COUNT
+    N_block_id = pid1
+
+    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)
+
+    if end_idx > start_idx:
+        M_block = tl.max_contiguous(start_idx + tl.arange(0, BLOCK_M), BLOCK_M)
+
+        K_block = K_block_id * BLOCK_K + tl.arange(0, BLOCK_K)
+        K_mask = K_block < K
+        K_block = tl.max_contiguous(tl.multiple_of(K_block % K, BLOCK_K), BLOCK_K)
+
+        N_block = N_block_id * BLOCK_N + tl.arange(0, BLOCK_N)
+        N_mask = N_block < N
+        N_block = tl.max_contiguous(tl.multiple_of(N_block % N, BLOCK_N), BLOCK_N)
+
+        M_idxs = M_block
+        xt_blk_ptrs = X_ptr + K_block[:, None] * stride_xn + M_idxs[None, :] * stride_xm
+        dy_blk_ptrs = DY_ptr + M_idxs[:, None] * stride_dym + N_block[None, :] * stride_dyk
+
+        acc = tl.zeros((BLOCK_K, BLOCK_N), dtype=ACC_TYPE)
+        iters = tl.cdiv(end_idx - start_idx, BLOCK_M)
+        for i in range(0, iters):
+            M_mask = (i * BLOCK_M + M_block) < end_idx
+            if NO_K_MASK:
+                xt = tl.load(xt_blk_ptrs, mask=M_mask[None, :])
+            else:
+                xt = tl.load(xt_blk_ptrs, mask=K_mask[:, None] & M_mask[None, :])
+            if NO_N_MASK:
+                dy = tl.load(dy_blk_ptrs, mask=M_mask[:, None])
+            else:
+                dy = tl.load(dy_blk_ptrs, mask=M_mask[:, None] & N_mask[None, :])
+            acc += tl.dot(xt, dy, out_dtype=ACC_TYPE, allow_tf32=allow_tf32)
+            xt_blk_ptrs += BLOCK_M * stride_xm
+            dy_blk_ptrs += BLOCK_M * stride_dym
+
+
+        DW_blk_ptrs = DW_ptr + E_idx * stride_dwe + K_block[:, None] * stride_dwk + N_block[None, :] * stride_dwn
+        acc = acc.to(DW_blk_ptrs.dtype.element_ty)
+        tl.store(DW_blk_ptrs, acc, mask=K_mask[:, None] & N_mask[None, :])
+
+
+def _config_grouping():
+    return [
+        triton.Config({'BLOCK_N': 256, 'BLOCK_K': 128}, num_stages=4, num_warps=4),
+        triton.Config({'BLOCK_N': 128, 'BLOCK_K': 64}, num_stages=4, num_warps=4),
+        triton.Config({'BLOCK_N': 64, 'BLOCK_K': 32}, num_stages=4, num_warps=4),
+    ]
+
+def group(A, sorted_expert_idxs, coeff=None, fan_out=1, out=None):
+    N = sorted_expert_idxs.size(0)
+    K = A.size(1)
+    assert A.size(0) * fan_out == N
+    if out is not None:
+        Y = out
+    else:
+        Y = torch.empty((N, K), dtype=A.dtype, device=A.device)
+        # print("grp init:", Y.size())
+    def grid(META):
+        grid_num = (triton.cdiv(META['N'], META['BLOCK_N']),)
+        return grid_num
+    _group[grid](
+        # A_ptr, stride_an, stride_ai,
+        A, A.stride(0), A.stride(1), coeff is not None, coeff, fan_out,
+        # Y_ptr, stride_yn, stride_yk,
+        Y, Y.stride(0), Y.stride(1),
+        # grouped_idx_ptr,
+        sorted_expert_idxs,
+        # N: tl.constexpr, K: tl.constexpr,
+        N, K
+    )
+    return Y
+
+@triton.autotune(configs=_config_grouping(), key=['K'])
+@triton.heuristics({
+    "NO_K_MASK": lambda args: (args['K'] % args['BLOCK_K']) == 0
+})
+@triton.jit
+def _group(
+    src_ptr, stride_sn, stride_sk, has_coeff: tl.constexpr, coeff_ptr, FAN_OUT: tl.constexpr,
+    tgt_ptr, stride_tn, stride_ti,
+    grouped_idx_ptr,
+    N: tl.constexpr, K: tl.constexpr,
+    BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
+    NO_K_MASK: tl.constexpr
+):
+    pid = tl.program_id(axis=0)
+
+    N_block_id = pid
+    N_blk = N_block_id * BLOCK_N + tl.arange(0, BLOCK_N)
+    N_mask = N_blk < N
+    N_blk = tl.max_contiguous(tl.multiple_of(N_blk % N, BLOCK_N), BLOCK_N)
+    N_idx = tl.load(grouped_idx_ptr + N_blk, mask=N_mask, other=0)
+
+    K_blk = tl.arange(0, BLOCK_K)
+    src_blk_ptrs = src_ptr + (N_idx // FAN_OUT)[:, None] * stride_sn + K_blk[None, :] * stride_sk
+    tgt_blk_ptrs = tgt_ptr + N_blk[:, None] * stride_tn + K_blk[None, :] * stride_ti
+
+    if has_coeff:
+        c = tl.load(coeff_ptr + N_idx, mask=N_mask)[:, None]
+
+    iters = tl.cdiv(K, BLOCK_K)
+    for i in range(0, iters):
+        if NO_K_MASK:
+            block = tl.load(src_blk_ptrs) # , mask=N_mask[:, None])
+            if has_coeff:
+                block *= c
+            tl.store(tgt_blk_ptrs, block, mask=N_mask[:, None])
+
+        else:
+            K_mask = (i * BLOCK_K + K_blk) < K
+            mask = N_mask[:, None] & K_mask[None, :]
+            block = tl.load(src_blk_ptrs, mask=mask)
+            if has_coeff:
+                block *= c
+            tl.store(tgt_blk_ptrs, block, mask=mask)
+
+        src_blk_ptrs += BLOCK_K * stride_sk
+        tgt_blk_ptrs += BLOCK_K * stride_ti

src/axolotl/monkeypatch/moe/single.py

@@ -0,0 +1,66 @@
+"""
+Adapted from:
+https://github.com/shawntan/scattermoe
+https://arxiv.org/abs/2403.08245
+"""
+
+import torch
+import triton
+import triton.language as tl
+from torch.nn import functional as F
+
+@triton.jit
+def _single2scatter(
+    X_ptr, stride_xm, stride_xk,
+    W_ptr, stride_we, stride_wk, stride_wn,
+    Y_ptr, stride_ym, stride_yn,
+    expert_idxs_ptr,
+    FAN_OUT: tl.constexpr,
+    K: tl.constexpr, N: tl.constexpr, E: tl.constexpr,
+    BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
+    ACC_TYPE: tl.constexpr,
+):
+    pid0 = tl.program_id(axis=0)
+    pid1 = tl.program_id(axis=1)
+
+    N_block_id = pid0
+    if FAN_OUT == 1:
+        in_idx = pid1
+    else:
+        in_idx = 0
+    out_idx = pid1
+
+    K_block = tl.arange(0, BLOCK_K)
+    N_block = tl.max_contiguous(tl.multiple_of((N_block_id * BLOCK_N + tl.arange(0, BLOCK_N)) % N, BLOCK_N), BLOCK_N)
+    E_idx = tl.load(expert_idxs_ptr + pid1)
+    X_blk_ptrs = X_ptr + in_idx * stride_xm + K_block[:, None] * stride_xk
+    W_blk_ptrs = W_ptr + E_idx * stride_we + K_block[:, None] * stride_wk + N_block[None, :] * stride_wn
+    acc = tl.zeros((1, BLOCK_N), dtype=ACC_TYPE)
+    for K_block_id in range(0, tl.cdiv(K, BLOCK_K)):
+        x = tl.load(X_blk_ptrs)
+        w = tl.load(W_blk_ptrs)
+        acc += tl.sum(x * w, axis=0)[None, :]
+        X_blk_ptrs += BLOCK_K * stride_xk
+        W_blk_ptrs += BLOCK_K * stride_wk
+    Y_blk_ptrs = Y_ptr + out_idx * stride_ym + N_block[None, :] * stride_yn
+    tl.store(Y_blk_ptrs, acc)
+
+def single2scatter(X, W, expert_idxs):
+    E, xdim, ydim = W.size()
+    k = expert_idxs.size(1)
+    assert X.size(0) == k or X.size(0) == 1
+    Y = torch.empty((k, ydim), device=X.device, dtype=X.dtype)
+    BLOCK_N = 128
+    BLOCK_K = 128
+    grid = ydim // BLOCK_N, k
+    _single2scatter[grid](
+        X, X.stride(0), X.stride(1),
+        W, W.stride(0), W.stride(1), W.stride(2),
+        Y, Y.stride(0), Y.stride(1),
+        expert_idxs,
+        FAN_OUT=Y.size(0) // X.size(0),
+        K=xdim, N=ydim, E=E,
+        BLOCK_N=BLOCK_N, BLOCK_K=BLOCK_K,
+        ACC_TYPE=tl.float32
+    )
+    return Y

src/axolotl/monkeypatch/multipack.py

@@ -1,6 +1,9 @@
 """multipack patching for v2 of sample packing"""
+import importlib
 
 import transformers
+from accelerate import init_empty_weights
+from transformers import AutoConfig, AutoModelForCausalLM
 from transformers.integrations import is_deepspeed_zero3_enabled
 
 from axolotl.monkeypatch.mixtral import patch_mixtral_moe_forward_zero3
@@ -12,11 +15,12 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "falcon",
     "phi",
     "gemma",
+    "gemmoe",
     "starcoder2",
 ]
 
 
-def patch_for_multipack(model_type):
+def patch_for_multipack(model_type, model_name=None):
     if model_type == "mixtral":
         transformers.models.mixtral.modeling_mixtral._get_unpad_data = (  # pylint: disable=protected-access
             get_unpad_data
@@ -43,3 +47,15 @@ def patch_for_multipack(model_type):
         transformers.models.starcoder2.modeling_starcoder2._get_unpad_data = (  # pylint: disable=protected-access
             get_unpad_data
         )
+    elif model_type == "gemmoe":
+        model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
+        # we need to load the model here in order for modeling_gemmoe to be available
+        with init_empty_weights():
+            AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
+        module_name = model_config.__class__.__module__.replace(
+            ".configuration_gemmoe", ".modeling_gemmoe"
+        )
+        modeling_gemmoe = importlib.import_module(module_name)
+        modeling_gemmoe._get_unpad_data = (  # pylint: disable=protected-access
+            get_unpad_data
+        )

src/axolotl/prompt_strategies/dpo/chatml.py

@@ -24,6 +24,25 @@ def argilla(
     return transform_fn
 
 
+def argilla_chat(
+    cfg,
+    **kwargs,
+):  # pylint: disable=possibly-unused-variable,unused-argument
+    """
+    for argilla/dpo-mix-7k conversations
+    """
+
+    def transform_fn(sample):
+        sample[
+            "prompt"
+        ] = f"<|im_start|>user\n{sample['chosen'][0]['content']}<|im_end|>\n<|im_start|>assistant\n"
+        sample["chosen"] = f"{sample['chosen'][1]['content']}<|im_end|>"
+        sample["rejected"] = f"{sample['rejected'][1]['content']}<|im_end|>"
+        return sample
+
+    return transform_fn
+
+
 def icr(
     cfg,
     **kwargs,

src/axolotl/prompt_strategies/sharegpt.py

@@ -1,10 +1,15 @@
 """Module containing the SimpleShareGPTPromptTokenizingStrategy class"""
+
 from typing import Any, Dict, Optional
 
 from fastchat.conversation import Conversation, SeparatorStyle, register_conv_template
 
 from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
 from axolotl.prompters import ShareGPTPrompterV2
+from axolotl.utils.tokenization import (
+    chatml_to_conversation,
+    merge_consecutive_messages,
+)
 
 
 def register_chatml_template(system_message=None):
@@ -19,6 +24,16 @@ def register_chatml_template(system_message=None):
             sep="<|im_end|>",
         )
     )
+    register_conv_template(
+        Conversation(
+            name="chatml_glaive",
+            system_template="<|im_start|>system\n{system_message}",
+            system_message=system_message,
+            roles=["<|im_start|>user", "<|im_start|>assistant", "<|im_start|>tool"],
+            sep_style=SeparatorStyle.CHATML,
+            sep="<|im_end|>",
+        )
+    )
 
 
 def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
@@ -77,6 +92,20 @@ def load_guanaco(tokenizer, cfg):
     )
 
 
+def load_glaive(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
+    conversation = (
+        ds_cfg["conversation"]
+        if ds_cfg and "conversation" in ds_cfg
+        else "chatml_glaive"
+    )
+    return GlaiveShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompterV2(conversation=conversation),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
 class SimpleShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
     """
     basic sharegpt strategy to grab conversations from the sample row
@@ -158,3 +187,15 @@ class UltrachatShareGPTPromptTokenizingStrategy(SimpleShareGPTPromptTokenizingSt
             {"from": role_map[t["role"]], "value": t["content"]} for t in conversations
         ]
         return turns
+
+
+class GlaiveShareGPTPromptTokenizingStrategy(SimpleShareGPTPromptTokenizingStrategy):
+    """
+    sharegpt strategy that remaps glaive data to sharegpt format
+    """
+
+    def get_conversation_thread(self, prompt):
+        conversation = chatml_to_conversation(prompt)
+        conversation = merge_consecutive_messages(conversation)
+
+        return conversation

src/axolotl/prompt_tokenizers.py

@@ -360,11 +360,19 @@ class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
                     LOG.warning(f"expected tuple, got {part}")
                     continue
 
-                user, assistant = conversation.roles
+                tool_role_label = None
+                if len(conversation.roles) == 3:
+                    (
+                        user_role_label,
+                        assistant_role_label,
+                        tool_role_label,
+                    ) = conversation.roles
+                else:
+                    user_role_label, assistant_role_label = conversation.roles
                 role, content = part
 
                 # Uses "in" because role contains extra characters
-                if user in role:
+                if user_role_label in role:
                     role = (
                         role.replace(role_remap[0]["from"], role_remap[0]["to"])
                         if role_remap
@@ -384,7 +392,7 @@ class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
                     else:
                         # everything from this is masked out from the labels
                         labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
-                elif assistant in role:
+                elif assistant_role_label in role:
                     role = (
                         role.replace(role_remap[1]["from"], role_remap[1]["to"])
                         if role_remap
@@ -426,6 +434,8 @@ class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
                     else:
                         # everything from this is masked out from the labels
                         labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+                elif tool_role_label and tool_role_label in role:
+                    labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
                 else:
                     LOG.warning(f"unhandled role: {role}")
                     continue

src/axolotl/prompters.py

@@ -267,6 +267,8 @@ class ShareGPTPrompter(Prompter):  # pylint: disable=too-few-public-methods
 
     role_key_human = "human"
     role_key_model = "gpt"
+    # Optional, only used for tool usage datasets.
+    role_key_tool = None
 
     def __init__(
         self,
@@ -274,6 +276,7 @@ class ShareGPTPrompter(Prompter):  # pylint: disable=too-few-public-methods
         conversation: Optional[Union[str, Conversation]] = None,
         role_key_human: Optional[str] = None,
         role_key_model: Optional[str] = None,
+        role_key_tool: Optional[str] = None,
     ):
         if conversation:
             if isinstance(conversation, Conversation):
@@ -286,6 +289,8 @@ class ShareGPTPrompter(Prompter):  # pylint: disable=too-few-public-methods
             self.role_key_human = role_key_human
         if role_key_model:
             self.role_key_model = role_key_model
+        if role_key_tool:
+            self.role_key_tool = role_key_tool
 
     def _build_result(self, source):
         if len(source) < 2:
@@ -303,6 +308,8 @@ class ShareGPTPrompter(Prompter):  # pylint: disable=too-few-public-methods
             source.pop(0)
 
         roles = {self.role_key_human: conv.roles[0], self.role_key_model: conv.roles[1]}
+        if self.role_key_tool:
+            roles[self.role_key_tool] = conv.roles[2]
 
         try:
             # Apply prompt templates

src/axolotl/train.py

@@ -19,7 +19,7 @@ from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
 from axolotl.common.cli import TrainerCliArgs
 from axolotl.logging_config import configure_logging
 from axolotl.utils.dict import DictDefault
-from axolotl.utils.freeze import freeze_parameters_except
+from axolotl.utils.freeze import freeze_layers_except
 from axolotl.utils.models import load_model, load_tokenizer
 from axolotl.utils.trainer import setup_trainer
 
@@ -99,7 +99,7 @@ def train(
     safe_serialization = cfg.save_safetensors is True
 
     if cfg.unfrozen_parameters:
-        freeze_parameters_except(model, cfg.unfrozen_parameters)
+        freeze_layers_except(model, cfg.unfrozen_parameters)
 
     trainer = setup_trainer(
         cfg,

src/axolotl/utils/bench.py

@@ -24,9 +24,9 @@ def check_cuda_device(default_value):
                 or not torch.cuda.is_available()
                 or device == "auto"
                 or torch.device(device).type == "cpu"
+                or torch.device(device).type == "meta"
             ):
                 return default_value
-
             return func(*args, **kwargs)
 
         return wrapper

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -1,6 +1,7 @@
 """
 Module for pydantic models for configuration
 """
+# pylint: disable=too-many-lines
 
 import logging
 import os
@@ -128,8 +129,10 @@ class RLType(str, Enum):
 class ChatTemplate(str, Enum):
     """Chat templates configuration subset"""
 
+    alpaca = "alpaca"  # pylint: disable=invalid-name
     chatml = "chatml"  # pylint: disable=invalid-name
     inst = "inst"  # pylint: disable=invalid-name
+    gemma = "gemma"  # pylint: disable=invalid-name
 
 
 class LoftQConfig(BaseModel):
@@ -179,6 +182,7 @@ class LoraConfig(BaseModel):
     peft_layers_to_transform: Optional[List[int]] = None
     peft: Optional[PeftConfig] = None
     peft_use_dora: Optional[bool] = None
+    peft_use_relora: Optional[bool] = None
 
     lora_on_cpu: Optional[bool] = None
     gptq: Optional[bool] = None
@@ -511,10 +515,12 @@ class AxolotlInputConfig(
 
     neftune_noise_alpha: Optional[float] = None
 
-    max_memory: Optional[Union[int, str]] = None
+    max_memory: Optional[
+        Dict[Union[int, Literal["cpu", "disk"]], Union[int, str]]
+    ] = None
     gpu_memory_limit: Optional[Union[int, str]] = None
 
-    chat_template: Optional[Union[Literal["chatml", "inst"], ChatTemplate]] = None
+    chat_template: Optional[ChatTemplate] = None
     default_system_message: Optional[str] = None
 
     # INTERNALS - document for now, generally not set externally
@@ -989,3 +995,10 @@ class AxolotlConfigWCapabilities(AxolotlInputConfig):
             )
 
         return data
+
+    @model_validator(mode="before")
+    @classmethod
+    def check_fsdp_deepspeed(cls, data):
+        if data.get("deepspeed") and data.get("fsdp"):
+            raise ValueError("deepspeed and fsdp cannot be used together.")
+        return data

src/axolotl/utils/freeze.py

@@ -3,13 +3,14 @@ module to freeze/unfreeze parameters by name
 """
 import logging
 import re
+from typing import Callable, List, Tuple
 
 from axolotl.utils.distributed import is_main_process
 
 LOG = logging.getLogger("axolotl.utils.freeze")
 
 
-def freeze_parameters_except(model, regex_patterns):
+def freeze_layers_except(model, regex_patterns):
     """
     Freezes all layers of the given model except for the layers that match given regex patterns.
     Periods in the patterns are treated as literal periods, not as wildcard characters.
@@ -17,22 +18,209 @@ def freeze_parameters_except(model, regex_patterns):
     Parameters:
     - model (nn.Module): The PyTorch model to be modified.
     - regex_patterns (list of str): List of regex patterns to match layer names to keep unfrozen.
+      Note that you cannot use a dot as a wildcard character in the patterns since it is reserved for separating layer names.
+      Also, to match the entire layer name, the pattern should start with "^" and end with "$", otherwise it will match any part of the layer name.
+      The range pattern part is optional and it is not compiled as a regex pattern which means you must put "$" before the range pattern if you want to match the entire layer name.
+      E.g., ["^model.embed_tokens.weight$[:32000]", "layers.2[0-9]+.block_sparse_moe.gate.[a-z]+$"]
 
     Returns:
     None; the model is modified in place.
     """
-    # Escape periods and compile the regex patterns
-    compiled_patterns = [
-        re.compile(pattern.replace(".", "\\.")) for pattern in regex_patterns
-    ]
+    if isinstance(regex_patterns, str):
+        regex_patterns = [regex_patterns]
 
-    # First, freeze all parameters in the model
-    for param in model.parameters():
-        param.requires_grad = False
+    patterns = [LayerNamePattern(pattern) for pattern in regex_patterns]
 
     # Unfreeze layers that match the regex patterns
     for name, param in model.named_parameters():
-        if any(pattern.match(name) for pattern in compiled_patterns):
-            if is_main_process():
-                LOG.debug(f"unfreezing {name}")
+        param.requires_grad = False
+        unfrozen_ranges = []
+        for pattern in patterns:
+            if not pattern.match(name):
+                continue
+
             param.requires_grad = True
+
+            if pattern.range is not None:
+                unfrozen_ranges.append(pattern.range)
+
+        merged_unfrozen_ranges = _merge_ranges(unfrozen_ranges, len(param))
+
+        if param.requires_grad and is_main_process():
+            unfrozen_ranges = (
+                f" with ranges {merged_unfrozen_ranges}"
+                if merged_unfrozen_ranges
+                else ""
+            )
+            LOG.debug(f"Unfrozen {name}{unfrozen_ranges}")
+
+        if not merged_unfrozen_ranges:
+            continue
+
+        # The range list we need is actually the inverted of the merged ranges
+        ranges_to_freeze = _invert_ranges(merged_unfrozen_ranges, len(param))
+
+        param.register_hook(_create_freeze_parameters_hook(ranges_to_freeze))
+
+    if is_main_process() and all(
+        not param.requires_grad for param in model.parameters()
+    ):
+        LOG.warning("All parameters are frozen. Model will not be trained.")
+
+
+def _invert_ranges(
+    given_ranges: List[Tuple[int, int]], layer_size: int
+) -> List[Tuple[int, int]]:
+    """
+    Inverts a list of ranges to obtain the ranges not covered by the given ranges.
+
+    Parameters:
+    - given_ranges (List[Tuple[int, int]]): List of ranges to invert. Each range is represented as a tuple of start (inclusive) and end (exclusive) indices.
+    - layer_size (int): The length of the layer. E.g., len(model.layer.weight)
+    Returns:
+    - List[Tuple[int, int]]: List of inverted ranges, where each range is represented as a tuple of start (inclusive) and end (exclusive) indices.
+    """
+    if not given_ranges:
+        return [(0, layer_size)]
+
+    inverted_ranges = []
+    current_start = 0
+
+    for start, end in sorted(given_ranges):
+        if start > current_start:
+            inverted_ranges.append((current_start, start))
+        current_start = max(current_start, end)
+
+    # Handle the case where the last given range does not reach the end of the total_size
+    if current_start < layer_size:
+        inverted_ranges.append((current_start, layer_size))
+
+    return inverted_ranges
+
+
+def _merge_ranges(
+    given_ranges: List[Tuple[int, int | None]], layer_size: int
+) -> List[Tuple[int, int]]:
+    """
+    Merges overlapping ranges and sorts the given ranges.
+
+    This function takes a list of ranges and merges any overlapping ranges. The ranges are represented
+    as tuples, where the first element is the start index (inclusive) and the second element is the end
+    index (exclusive). The end index can be None, indicating that the range extends to the end of the
+    sequence.
+
+    Parameters:
+    - given_ranges (List[Tuple[int, int | None]]): List of ranges to merge.
+    - layer_size (int): The length of the layer. E.g., len(model.layer.weight)
+
+    Returns:
+    - List[Tuple[int, int]]: List of merged ranges, as start (inclusive) and end (exclusive) indices.
+    """
+    # End of each range can be determined now since we have the total size
+    processed_ranges = [
+        (start, end if end is not None else layer_size) for start, end in given_ranges
+    ]
+
+    # No need to merge if there's only one or no ranges
+    if len(processed_ranges) <= 1:
+        return processed_ranges
+
+    sorted_ranges = sorted(processed_ranges)
+
+    merged_ranges = [sorted_ranges[0]]
+    for start, end in sorted_ranges[1:]:
+        prev_start, prev_end = merged_ranges[-1]
+        if start <= prev_end:
+            merged_ranges[-1] = (prev_start, max(prev_end, end))
+        else:
+            merged_ranges.append((start, end))
+
+    return merged_ranges
+
+
+def _create_freeze_parameters_hook(ranges_to_freeze: List[Tuple[int, int]]) -> Callable:
+    """
+    Create a hook to freeze parameters in specified ranges by setting their gradients to zero.
+
+    This function takes a list of tuples representing the ranges of indices to freeze. Each tuple should contain
+    two integers representing the start and end indices of the range.
+
+    Parameters:
+    - ranges_to_freeze (List[Tuple[int, int]]): Ranges of indices to freeze.
+
+    Returns:
+    - Callable: A hook function to be used with `register_hook` on parameters.
+
+    Example usage:
+    ```
+    ranges_to_freeze = [(0, 10), (20, 30)]
+    hook = _create_freeze_parameters_hook(ranges_to_freeze)
+    model.register_hook(hook)
+    ```
+    """
+
+    def freeze_parameters_hook(gradients):
+        for start, end in ranges_to_freeze:
+            gradients[start:end].zero_()
+
+    return freeze_parameters_hook
+
+
+class LayerNamePattern:
+    """
+    Represents a regex pattern for layer names, potentially including a parameter index range.
+    """
+
+    def __init__(self, pattern: str):
+        """
+        Initializes a new instance of the LayerNamePattern class.
+
+        Parameters:
+        - pattern (str): The regex pattern for layer names, potentially including a parameter index range.
+        """
+        self.raw_pattern = pattern
+        name_pattern, self.range = self._parse_pattern(pattern)
+        self.name_regex = re.compile(name_pattern.replace(".", "\\."))
+
+    def match(self, name: str) -> bool:
+        """
+        Checks if the given layer name matches the regex pattern.
+
+        Parameters:
+        - name (str): The layer name to check.
+
+        Returns:
+        - bool: True if the layer name matches the pattern, False otherwise.
+        """
+        return self.name_regex.match(name) is not None
+
+    def _parse_pattern(self, pattern: str) -> Tuple[str, Tuple[int, int | None] | None]:
+        """
+        Extracts the range pattern from the given pattern.
+
+        Parameters:
+        - pattern (str): The pattern to extract the range from.
+
+        Returns:
+        - Tuple[str, Tuple[int, int | None] | None]: A tuple containing the regex pattern to match the layer name without the range pattern and the range of layer indices to match, if specified.
+        """
+        match = re.match(r"^(.+)\[([0-9]*)(?::([0-9]*))?\]$", pattern)
+        if not match:
+            return pattern, None
+
+        base_pattern, start_part, end_part = match.groups()
+
+        if end_part is None and start_part.isdecimal():
+            index = int(start_part)
+            return base_pattern, (index, index + 1)
+
+        # [:end] or [start:] or [start:end]
+        start = int(start_part) if start_part else 0
+        end = int(end_part) if end_part else None
+
+        if end is not None and start >= end:
+            raise ValueError(
+                f"Invalid range in layer name pattern: {pattern}."
+                "End of range must be greater than start."
+            )
+        return base_pattern, (start, end)

src/axolotl/utils/models.py

@@ -1,13 +1,20 @@
 """Module for models and model loading"""
+# pylint: disable=too-many-lines
+
 import logging
 import math
 import os
-from typing import Any, Dict, Optional, Tuple, Union  # noqa: F401
+import types
+from typing import Any, Dict, List, Optional, Tuple, Type, Union  # noqa: F401
 
 import addict
 import bitsandbytes as bnb
+import safetensors
 import torch
 import transformers
+from accelerate import init_empty_weights
+from bitsandbytes.nn import Linear4bit, Params4bit
+from fastcore.parallel import parallel
 from peft import (
     LoftQConfig,
     PeftConfig,
@@ -16,6 +23,7 @@ from peft import (
     prepare_model_for_kbit_training,
 )
 from peft.tuners.lora import QuantLinear
+from torch import Tensor, nn
 from transformers import (  # noqa: F401
     AddedToken,
     AutoConfig,
@@ -27,7 +35,9 @@ from transformers import (  # noqa: F401
     PreTrainedTokenizerBase,
 )
 from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
+from transformers.utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, hub
 
+from axolotl.core.policies.auto_wrap import SUPPORTED_AUTO_WRAP_MODEL_TYPES
 from axolotl.models.mamba import fix_mamba_attn_for_loss
 from axolotl.monkeypatch.multipack import (
     SUPPORTED_MULTIPACK_MODEL_TYPES,
@@ -262,6 +272,117 @@ def load_tokenizer(cfg):
     return tokenizer
 
 
+def replace_linear(
+    model: nn.Module,
+    linear_replacement: Type[nn.Module],
+    quant_config: Union[dict, None] = None,
+    skip_modules=None,
+    **kwargs,
+):
+    """
+    Replace linear modules with a new Linear module.
+    Parameters:
+        model (`torch.nn.Module`):
+            Input model or `torch.nn.Module` as the function is run recursively.
+        linear_replacement (`torch.nn.Module`):
+            The linear module that replaces the old one. Only expects standard arguments.
+            If other arguments need to be passed, use a lambda.
+        skip_modules (`List[str]`, *optional*, defaults to `lm_head`):
+            List of modules names not to convert. Defaults to `lm_head`.
+    """
+    if skip_modules is None:
+        skip_modules = ["lm_head"]
+    for name, module in model.named_children():
+        if len(list(module.children())) > 0:
+            replace_linear(
+                module, linear_replacement, quant_config, skip_modules, **kwargs
+            )
+
+        if isinstance(module, torch.nn.Linear) and name not in skip_modules:
+            if issubclass(linear_replacement, Linear4bit):
+                model._modules[  # pylint: disable=protected-access
+                    name
+                ] = linear_replacement(
+                    module.in_features,
+                    module.out_features,
+                    module.bias is not None,
+                    **kwargs,
+                )
+            else:
+                raise ValueError(
+                    f"Unsupported linear replacement: {type(linear_replacement)}"
+                )
+    return model
+
+
+def load_and_quantize(
+    module: nn.Module,
+    name: str,
+    value: Tensor,
+    device: torch.device = None,
+    dtype: torch.dtype = None,
+    skip_names: Optional[List[str]] = None,
+    is_meta_rank: bool = False,
+    low_memory: bool = True,
+    verbose: bool = False,
+    quant_method: str = "bnb",
+):
+    """
+    Loads `value` tensor into submodule of `module`, optionally skipping `skip_names` and converting to `dtype`.
+
+    Quantizes `Params4bit` on `device` then places on "cpu" if low_memory=True or "meta" if is_meta_rank=True.
+    """
+
+    if skip_names is None:
+        skip_names = []
+
+    def place_on_device(value):
+        if is_meta_rank:
+            device = "meta"
+        elif low_memory:
+            device = "cpu"
+        else:
+            device = "cuda"
+        return value.to(device=device, dtype=dtype)
+
+    if any(skip_name in name for skip_name in skip_names):
+        if verbose:
+            print(f"Skipping {name} because it is in skip_names")
+        return
+
+    module_key, _, value_key = name.rpartition(".")
+    try:
+        submodule = module.get_submodule(module_key)
+    except AttributeError as exc:
+        print(f"Module {module_key} not found:\n{exc}")
+        return
+
+    try:
+        if quant_method == "bnb":
+            param = submodule.get_parameter(value_key)
+            if isinstance(param, Params4bit):
+                # With `sync_module_states=True`, a meta device Params4bit needs to be the same
+                # shape as the quantized Params4bit with an initialized quant_state. However,
+                # FSDP only syncs parameters and buffers, so the quant_state isn't copied. This
+                # workaround quantizes Params4bit to initialize quant_state on all ranks, then
+                # replaces Params4bit's data with a meta tensor to free memory on non-rank 0.
+                value = type(param)(
+                    value.to(device=device, dtype=dtype).data, **param.__dict__
+                ).cuda(device)
+                if is_meta_rank:
+                    value = type(param)(value.data.to("meta"), **value.__dict__)
+                elif low_memory:
+                    value = type(param)(value.data.to("cpu"), **value.__dict__)
+            else:
+                value = type(param)(place_on_device(value).data)
+
+    except AttributeError:
+        # it's a buffer
+        value = place_on_device(value)
+
+    setattr(submodule, value_key, value)
+
+
 def load_model(
     cfg: DictDefault,
     tokenizer: PreTrainedTokenizerBase,
@@ -308,7 +429,7 @@ def load_model(
         and cfg.flash_attention
         and cfg.sample_packing
     ):
-        patch_for_multipack(cfg.model_config_type)
+        patch_for_multipack(cfg.model_config_type, model_name=cfg.base_model)
     elif cfg.is_llama_derived_model:
         # Modify all llama derived models in one block
 
@@ -394,7 +515,7 @@ def load_model(
 
     if max_memory is not None:
         # Based on https://github.com/togethercomputer/OpenChatKit/blob/main/inference/bot.py
-        from accelerate import infer_auto_device_map, init_empty_weights
+        from accelerate import infer_auto_device_map
 
         with init_empty_weights():
             model_canvas = AutoModelForCausalLM.from_config(model_config)
@@ -496,8 +617,78 @@ def load_model(
         model_kwargs["attn_implementation"] = "eager"
         model_config._attn_implementation = "eager"  # pylint: disable=protected-access
 
+    qlora_fsdp = (
+        cfg.fsdp
+        and cfg.adapter == "qlora"
+        and model_config.model_type in SUPPORTED_AUTO_WRAP_MODEL_TYPES
+    )
+
     try:
-        if (
+        if qlora_fsdp:
+            if cfg.bf16 or cfg.bfloat16:
+                torch_dtype, compute_dtype = torch.float32, torch.bfloat16
+            elif cfg.fp16 or cfg.float16:
+                torch_dtype, compute_dtype = torch.float32, torch.float16
+            else:
+                torch_dtype, compute_dtype = torch.float32, torch.float16
+
+            with init_empty_weights():
+                LOG.info("Loading model with empty weights.")
+                model = AutoModelForCausalLM.from_config(model_config)
+                model.model = replace_linear(
+                    model.model,
+                    Linear4bit,
+                    compute_dtype=compute_dtype,
+                    quant_type="nf4",
+                    quant_storage=torch_dtype,
+                )
+
+            model.is_loaded_in_4bit = True
+
+            # Grab the safetensors files that hold the weights
+            try:
+                idx = hub.cached_file(base_model, SAFE_WEIGHTS_INDEX_NAME)
+                files, _ = hub.get_checkpoint_shard_files(base_model, idx)
+            except OSError:
+                try:
+                    # This means the model doesn't have a model.safetensors.index.json because it is not sharded
+                    files = []
+                    files.append(hub.cached_file(base_model, SAFE_WEIGHTS_NAME))
+                except OSError as exc:
+                    # This means the model probably doesn't have a safetensors file
+                    raise exc
+
+            # Load in the weights, using our custom load_and_quantize method which quantizes Params4bit on the fly
+            # and then places each layer on CPU or meta if using low_memory to minimize GPU memory usage
+            def load_and_quantize_parallel(name_param, model, **kwargs):
+                name, param = name_param
+                load_and_quantize(model, name, param, **kwargs)
+
+            param_count = sum((p.numel() for n, p in model.named_parameters()))
+            for filename in files:
+                weights = safetensors.torch.load_file(filename)
+                quant_method = "bnb"
+                devprops = torch.cuda.get_device_properties(torch.cuda.current_device())
+                left = int(os.cpu_count() / torch.cuda.device_count())
+                right = int(
+                    8 * (devprops.total_memory / 1e9 / 40) * (70 / (param_count / 1e9))
+                )
+                n_workers = min(left, right)
+                parallel(
+                    load_and_quantize_parallel,
+                    weights.items(),
+                    n_workers=n_workers,
+                    threadpool=True,
+                    model=model,
+                    dtype=torch_dtype,
+                    device=cfg.local_rank,
+                    skip_names=[],
+                    is_meta_rank=(cfg.local_rank != 0),
+                    verbose=False,
+                    quant_method=quant_method,
+                )
+
+        elif (
             model_config.model_type == "llama"
             and not cfg.trust_remote_code
             and not cfg.gptq
@@ -524,32 +715,27 @@ def load_model(
                 if cfg.flash_attn_fuse_qkv:
                     LOG.info("patching with fused QKV")
                     replace_llama_qkv_with_fused(model)
-        # elif model_type == "GPTNeoXForCausalLM" and cfg.flash_attention:
-        #     This is a WIP, still an issue with the backward pass
-        #     RuntimeError: grad can be implicitly created only for scalar outputs
-        #     TODO: try config.sequence_parallel = False
-        #     # https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/tests/models/test_gpt_neox.py#L12
-        #     # https://github.com/HazyResearch/flash-attention/tree/main/training#model-components
-        #     # add `**kwargs` to https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/flash_attn/models/gpt.py#L442
-        #     from flash_attn.utils.pretrained import state_dict_from_pretrained
-        #     from flash_attn.models.gpt import GPTLMHeadModel
-        #     from flash_attn.models.gpt_neox import remap_state_dict_hf_gpt_neox, gpt_neox_config_to_gpt2_config
-        #     from transformers import GPTNeoXConfig
-        #     config = gpt_neox_config_to_gpt2_config(GPTNeoXConfig.from_pretrained(base_model))
-        #     config.use_flash_attn = True
-        #     config.fused_bias_fc = True
-        #     config.fused_mlp = True  # GPT-NeoX-20B uses "gelu_fast"
-        #     config.activation_function = "gelu_fast"
-        #     config.fused_dropout_add_ln = True
-        #     # config.residual_in_fp32 = True
-        #
-        #     model: GPTLMHeadModel = GPTLMHeadModel.from_pretrained(
-        #         base_model,
-        #         config,
-        #         dtype=torch_dtype,
-        #         device=cfg.device,
-        #     )
-        #     model.train() # sets to train instead of eval mode
+        elif (
+            model_config.model_type == "mixtral"
+            and not cfg.adapter
+            and cfg.fuse_moe
+        ):
+            from axolotl.monkeypatch.utils import set_module_name
+            from axolotl.monkeypatch.moe.moe import SparseMoeBlock
+            from transformers.models.mixtral.modeling_mixtral import MixtralSparseMoeBlock
+
+            for name, module in model.named_modules():
+                if isinstance(module, MixtralSparseMoeBlock):
+                    smoe = SparseMoeBlock(
+                        experts=module.experts,
+                        gate=module.gate,
+                        hidden_dim=module.hidden_dim,
+                        ffn_dim=module.ffn_dim,
+                        num_experts=module.num_experts,
+                        top_k=module.top_k,
+                    )
+                    set_module_name(model, name, smoe)
+
         elif model_type == "MambaLMHeadModel":
             # FIXME this is janky at best and hacked together to make it work
             MambaLMHeadModel = fix_mamba_attn_for_loss()  # pylint: disable=invalid-name
@@ -613,7 +799,7 @@ def load_model(
         LOG.exception(err)
         raise err
 
-    if isinstance(model, (PeftModel, PeftModelForCausalLM)):
+    if isinstance(model, (PeftModel, PeftModelForCausalLM)) and not qlora_fsdp:
         model = model.merge_and_unload()
 
     embeddings_len = (
@@ -692,6 +878,9 @@ def load_model(
     if cfg.adapter == "lora" and loftq_bits:
         skip_prepare_model_for_kbit_training = True
 
+    if qlora_fsdp:
+        skip_prepare_model_for_kbit_training = True
+
     if cfg.adapter in ["lora", "qlora"]:
         if cfg.gradient_checkpointing:
             model.gradient_checkpointing_enable()
@@ -706,7 +895,7 @@ def load_model(
 
     # LlamaRMSNorm layers are in fp32 after kbit_training or full finetune, so we need to
     # convert them back to fp16/bf16 for flash-attn compatibility.
-    if needs_fa2_dtype or cfg.flash_attention:
+    if (needs_fa2_dtype or cfg.flash_attention) and not qlora_fsdp:
         LOG.info("converting modules to %s for flash attention", cfg.torch_dtype)
         for name, module in model.named_modules():
             if "norm" in name:
@@ -724,7 +913,12 @@ def load_model(
         else:
             model, lora_config = load_adapter(model, cfg, cfg.adapter)
 
-    if cfg.ddp and not load_in_8bit and not (cfg.rl and cfg.load_in_4bit):
+    if (
+        cfg.ddp
+        and not load_in_8bit
+        and not (cfg.rl and cfg.load_in_4bit)
+        and not qlora_fsdp
+    ):
         # TODO revaldate this conditional
         model.to(f"cuda:{cfg.local_rank}")
 
@@ -813,6 +1007,30 @@ def find_all_linear_names(model):
     return list(lora_module_names)
 
 
+def setup_quantized_meta_for_peft(model: nn.Module):
+    """Replaces `quant_state.to` with a dummy function to prevent PEFT from moving `quant_state` to meta device"""
+
+    def temp_to_method(self, *args, **kwargs):  # pylint: disable=unused-argument
+        return self
+
+    for param in model.parameters():
+        if isinstance(param, Params4bit):
+            param.quant_state._orig_to = (  # pylint: disable=protected-access
+                param.quant_state.to
+            )
+            param.quant_state.to = types.MethodType(temp_to_method, param.quant_state)
+
+
+def setup_quantized_peft_meta_for_training(model: nn.Module):
+    """Replaces dummy `quant_state.to` method with the original function to allow training to continue"""
+    for param in model.parameters():
+        if isinstance(param, Params4bit) and hasattr(param.quant_state, "_orig_to"):
+            param.quant_state.to = (
+                param.quant_state._orig_to  # pylint: disable=protected-access
+            )
+            param.quant_state._orig_to = None  # pylint: disable=protected-access
+
+
 def load_lora(model, cfg, inference=False, config_only=False):
     # type: (PreTrainedModel, DictDefault, bool, bool) -> Tuple[Optional[PreTrainedModel], Optional[PeftConfig]]
 
@@ -832,6 +1050,8 @@ def load_lora(model, cfg, inference=False, config_only=False):
         lora_config_kwargs["init_lora_weights"] = "loftq"
     if cfg.peft_use_dora:
         lora_config_kwargs["use_dora"] = cfg.peft_use_dora
+    if cfg.peft_use_rslora:
+        lora_config_kwargs["use_rslora"] = cfg.use_rslora
 
     lora_config = LoraConfig(
         r=cfg.lora_r,
@@ -849,6 +1069,11 @@ def load_lora(model, cfg, inference=False, config_only=False):
     if config_only:
         return None, lora_config
 
+    rank = int(os.environ.get("LOCAL_RANK", 0))
+
+    if cfg.fsdp and cfg.adapter == "qlora" and rank != 0:
+        setup_quantized_meta_for_peft(model)
+
     if cfg.lora_model_dir:
         LOG.debug("Loading pretrained PEFT - LoRA")
         model_kwargs: Any = {}
@@ -864,6 +1089,9 @@ def load_lora(model, cfg, inference=False, config_only=False):
     else:
         model = get_peft_model(model, lora_config)
 
-    model.print_trainable_parameters()
+    if rank == 0:
+        model.print_trainable_parameters()
+    elif cfg.fsdp and cfg.adapter == "qlora":
+        setup_quantized_peft_meta_for_training(model)
 
     return model, lora_config

src/axolotl/utils/tokenization.py

@@ -2,6 +2,8 @@
 
 
 import logging
+import re
+from typing import Dict, List
 
 from termcolor import colored
 
@@ -36,3 +38,65 @@ def check_example_labels(example, tokenizer, text_only=False):
     LOG.info("\n\n\n")
 
     return " ".join(colored_tokens)
+
+
+GLAIVE_ROLES = ["USER", "ASSISTANT", "FUNCTION RESPONSE"]
+GLAIVE_TO_SHAREGPT_ROLE = {
+    "SYSTEM": "system",
+    "USER": "human",
+    "ASSISTANT": "gpt",
+    "FUNCTION RESPONSE": "tool",
+}
+
+GLAIVE_MSG_REGEX = re.compile(rf"({'|'.join(GLAIVE_ROLES)}): ")
+
+
+def chatml_to_conversation(row: Dict[str, str]) -> List[Dict[str, str]]:
+    """
+    Converts a ChatML formatted row to a list of messages in ShareGPT format.
+    Initially based off https://github.com/lilacai/lilac/blob/main/notebooks/GlaiveToShareGPT.ipynb.
+    """
+
+    system_prompt = row.get("system")
+    if system_prompt:
+        system_prompt = system_prompt.removeprefix("SYSTEM: ")
+
+    chat_str = row["chat"]
+    chat_msgs = [s.strip() for s in GLAIVE_MSG_REGEX.split(chat_str) if s]
+
+    chat_msg_dicts = [
+        {"from": GLAIVE_TO_SHAREGPT_ROLE[role], "value": value}
+        for role, value in zip(chat_msgs[::2], chat_msgs[1::2])
+    ]
+
+    if system_prompt:
+        chat_msg_dicts = [
+            {"from": GLAIVE_TO_SHAREGPT_ROLE["SYSTEM"], "value": system_prompt}
+        ] + chat_msg_dicts
+
+    return chat_msg_dicts
+
+
+def merge_consecutive_messages(messages):
+    """
+    Merge consecutive messages from the same sender into a single message.
+    This can be useful with datasets that contain multiple consecutive tool calls.
+    """
+
+    merged_messages = []
+    current_from = None
+    current_message = ""
+
+    for msg in messages:
+        if current_from == msg["from"]:
+            current_message += msg["value"]
+        else:
+            if current_from is not None:
+                merged_messages.append({"from": current_from, "value": current_message})
+            current_from = msg["from"]
+            current_message = msg["value"]
+
+    if current_from is not None:
+        merged_messages.append({"from": current_from, "value": current_message})
+
+    return merged_messages

tests/monkeypatch/test_moe.py

@@ -0,0 +1,60 @@
+import torch
+import pytest
+from torch import nn
+from torch.nn import functional as F
+from axolotl.monkeypatch.moe.mlp import FusedExperts
+from axolotl.monkeypatch.moe.moe import SparseMoeBlock
+
+from transformers.models.mixtral.modeling_mixtral import MixtralSparseMoeBlock, MixtralConfig
+
+def test_fused_mixtral_moe():
+    # NOTE: Requires torch 2.2.0
+    # Set random seeds for reproducibility
+    torch.set_default_dtype(torch.float16)
+    torch.set_default_device("cuda")
+    torch.manual_seed(0)
+
+    # Define the configuration for the MixtralSparseMoeBlock
+    config = MixtralConfig(
+        hidden_size=128,
+        intermediate_size=512,
+        num_local_experts=8,
+        num_experts_per_tok=2,
+    )
+
+    # Initialize the MixtralSparseMoeBlock and SparseMoeBlock with the same configuration
+    mixtral_moe = MixtralSparseMoeBlock(config)
+    sparse_moe = SparseMoeBlock(
+        experts=mixtral_moe.experts,
+        gate=mixtral_moe.gate,
+        hidden_dim=config.hidden_size,
+        ffn_dim=config.intermediate_size,
+        num_experts=config.num_local_experts,
+        top_k=config.num_experts_per_tok
+    )
+
+    assert torch.cat([
+        mixtral_moe.experts[0].w1.weight.data,
+        mixtral_moe.experts[0].w3.weight.data], dim=0
+    ).equal(sparse_moe.experts.experts.weight[0])
+
+    # Generate random input data
+    batch_size = 16
+    sequence_length = 32
+    input_data = torch.randn(batch_size, sequence_length, config.hidden_size)
+
+    # Run the forward pass with gradients for both models
+    with torch.no_grad():
+        mixtral_output, mixtral_router_logits = mixtral_moe(input_data)
+        sparse_output, sparse_router_logits = sparse_moe(input_data)
+
+    # Compute the difference between the outputs
+    output_diff = torch.abs(mixtral_output - sparse_output).mean().item()
+    router_diff = torch.abs(mixtral_router_logits - sparse_router_logits).mean().item()
+
+    # Define the tolerance for the difference
+    tolerance = 0.05
+
+    # # Check if the difference is within the tolerance
+    assert output_diff < 0.05, f"Output difference is {output_diff}, which is greater than the tolerance of {tolerance}"
+    assert router_diff == 0, f"Output difference is {output_diff}, which is greater than the tolerance of {tolerance}"

tests/prompt_strategies/test_sharegpt.py

@@ -1,6 +1,7 @@
 """
 Test module for sharegpt integration w chatml
 """
+
 import pytest
 from datasets import Dataset
 from tokenizers import AddedToken
@@ -8,6 +9,7 @@ from transformers import AutoTokenizer
 
 from axolotl.datasets import TokenizedPromptDataset
 from axolotl.prompt_strategies.sharegpt import (
+    GlaiveShareGPTPromptTokenizingStrategy,
     SimpleShareGPTPromptTokenizingStrategy,
     register_chatml_template,
 )
@@ -48,6 +50,18 @@ def fixture_sharegpt_dataset():
     )
 
 
+@pytest.fixture(name="glaive_dataset")
+def fixture_sharegpt_glaive_dataset():
+    return Dataset.from_list(
+        [
+            {
+                "system": "SYSTEM: This is a system prompt",
+                "chat": "USER: Can you book a flight for me from New York to London? ASSISTANT: I'm sorry, but I don't have the capability to book flights.  <|endoftext|>",
+            }
+        ]
+    )
+
+
 @pytest.fixture(name="tokenizer")
 def fixture_tokenizer():
     tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
@@ -156,3 +170,29 @@ class TestSharegpt:
             32001, 13892, 13, 12684, 17664, 32000, 28705, 13,  # gpt
         ]
         # fmt: on
+
+    def test_chatml_glaive(self, glaive_dataset, tokenizer):
+        strategy = GlaiveShareGPTPromptTokenizingStrategy(
+            ShareGPTPrompterV2(
+                conversation="chatml",
+                role_key_model=None,
+                role_key_human=None,
+            ),
+            tokenizer,
+            True,  # train_on_inputs
+            2048,  # sequence_len
+        )
+
+        dataset_wrapper = TokenizedPromptDataset(
+            strategy, glaive_dataset, process_count=1
+        )
+
+        labels = dataset_wrapper[0]["labels"]
+        # fmt: off
+        assert labels == [
+            1,  # bos
+            32001, 1587, 13, 3260, 349, 264, 1587, 11510, 32000, 28705, 13,  # system
+            32001, 2188, 13, 6325, 368, 1820, 264, 9314, 354, 528, 477, 1450, 2726, 298, 4222, 28804, 32000, 28705, 13,  # human
+            32001, 13892, 13, 28737, 28742, 28719, 7371, 28725, 562, 315, 949, 28742, 28707, 506, 272, 21368, 298, 1820, 22447, 28723, 28705, 523, 28766, 416, 1009, 772, 28766, 28767, 32000, 28705, 13  # gpt
+        ]
+        # fmt: on

tests/test_freeze.py

@@ -0,0 +1,285 @@
+"""
+This module contains unit tests for the `freeze_layers_except` function.
+
+The `freeze_layers_except` function is used to freeze layers in a model, except for the specified layers.
+The unit tests in this module verify the behavior of the `freeze_layers_except` function in different scenarios.
+"""
+
+import unittest
+
+import torch
+from torch import nn
+
+from axolotl.utils.freeze import freeze_layers_except
+
+ZERO = [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
+ONE_TO_TEN = [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0]
+
+
+class TestFreezeLayersExcept(unittest.TestCase):
+    """
+    A test case class for the `freeze_layers_except` function.
+    """
+
+    def setUp(self):
+        self.model = _TestModel()
+
+    def test_freeze_layers_with_dots_in_name(self):
+        freeze_layers_except(self.model, ["features.layer"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+    def test_freeze_layers_without_dots_in_name(self):
+        freeze_layers_except(self.model, ["classifier"])
+        self.assertFalse(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertTrue(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+    def test_freeze_layers_regex_patterns(self):
+        # The second pattern cannot match because only characters 'a' to 'c' are allowed after the word 'class', whereas it should be matching the character 'i'.
+        freeze_layers_except(self.model, [r"^features.[a-z]+.weight$", r"class[a-c]+"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+    def test_all_layers_frozen(self):
+        freeze_layers_except(self.model, [])
+        self.assertFalse(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be frozen.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+    def test_all_layers_unfrozen(self):
+        freeze_layers_except(self.model, ["features.layer", "classifier"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertTrue(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be trainable.",
+        )
+
+    def test_freeze_layers_with_range_pattern_start_end(self):
+        freeze_layers_except(self.model, ["features.layer[1:5]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ZERO,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+            ]
+        )
+
+    def test_freeze_layers_with_range_pattern_single_index(self):
+        freeze_layers_except(self.model, ["features.layer[5]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [ZERO, ZERO, ZERO, ZERO, ZERO, ONE_TO_TEN, ZERO, ZERO, ZERO, ZERO]
+        )
+
+    def test_freeze_layers_with_range_pattern_start_omitted(self):
+        freeze_layers_except(self.model, ["features.layer[:5]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+            ]
+        )
+
+    def test_freeze_layers_with_range_pattern_end_omitted(self):
+        freeze_layers_except(self.model, ["features.layer[4:]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+            ]
+        )
+
+    def test_freeze_layers_with_range_pattern_merge_included(self):
+        freeze_layers_except(self.model, ["features.layer[4:]", "features.layer[5:6]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+            ]
+        )
+
+    def test_freeze_layers_with_range_pattern_merge_intersect(self):
+        freeze_layers_except(self.model, ["features.layer[4:7]", "features.layer[6:8]"])
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ONE_TO_TEN,
+                ZERO,
+                ZERO,
+            ]
+        )
+
+    def test_freeze_layers_with_range_pattern_merge_separate(self):
+        freeze_layers_except(
+            self.model,
+            ["features.layer[1:2]", "features.layer[3:4]", "features.layer[5:6]"],
+        )
+        self.assertTrue(
+            self.model.features.layer.weight.requires_grad,
+            "model.features.layer should be trainable.",
+        )
+        self.assertFalse(
+            self.model.classifier.weight.requires_grad,
+            "model.classifier should be frozen.",
+        )
+
+        self._assert_gradient_output(
+            [
+                ZERO,
+                ONE_TO_TEN,
+                ZERO,
+                ONE_TO_TEN,
+                ZERO,
+                ONE_TO_TEN,
+                ZERO,
+                ZERO,
+                ZERO,
+                ZERO,
+            ]
+        )
+
+    def _assert_gradient_output(self, expected):
+        input_tensor = torch.tensor([ONE_TO_TEN], dtype=torch.float32)
+
+        self.model.features.layer.weight.grad = None  # Reset gradients
+        output = self.model.features.layer(input_tensor)
+        loss = output.sum()
+        loss.backward()
+
+        expected_grads = torch.tensor(expected)
+        torch.testing.assert_close(
+            self.model.features.layer.weight.grad, expected_grads
+        )
+
+
+class _SubLayerModule(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.layer = nn.Linear(10, 10)
+
+
+class _TestModel(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.features = _SubLayerModule()
+        self.classifier = nn.Linear(10, 2)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_prompt_tokenizers.py

@@ -1,4 +1,5 @@
 """Module for testing prompt tokenizers."""
+
 import json
 import logging
 import unittest
@@ -18,6 +19,7 @@ from axolotl.prompt_strategies.llama2_chat import (
     Llama2ChatPrompter,
     LLama2ChatTokenizingStrategy,
 )
+from axolotl.prompt_strategies.sharegpt import GlaiveShareGPTPromptTokenizingStrategy
 from axolotl.prompt_tokenizers import (
     AlpacaPromptTokenizingStrategy,
     ShareGPTPromptTokenizingStrategy,
@@ -266,6 +268,23 @@ class TestPromptTokenizationStrategies(unittest.TestCase):
             idx = res["input_ids"].index(20255)  # assistant token
             assert res["labels"][idx] == -100
 
+    def test_glaive_tool_label_ignore(self):
+        conversation = {
+            "system": "SYSTEM: This is a system prompt",
+            "chat": "USER: Can you book a flight for me from New York to London? ASSISTANT: I'm sorry, but I don't have the capability to book flights.  <|endoftext|>",
+        }
+        prompter = ShareGPTPrompterV2()
+        strat = GlaiveShareGPTPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        with self._caplog.at_level(logging.WARNING):
+            res = strat.tokenize_prompt(conversation)
+            idx = res["input_ids"].index(13566)  # assistant token
+            assert res["labels"][idx] == -100
+
     def test_no_sys_prompt(self):
         """
         tests the interface between the user and assistant parts