Remove FP32 cast

* feat: add check for quantized model

* chore: refactor and add another check

* Update src/axolotl/utils/models.py

---------

Co-authored-by: Wing Lian <wing.lian@gmail.com>

README.md

@@ -612,6 +612,12 @@ eval_sample_packing:
 sample_packing_eff_est:
 total_num_tokens:
 
+# Passed through to transformers when loading the model when launched without accelerate
+# Use `sequential` when training w/ model parallelism to limit memory
+device_map:
+# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.
+max_memory:
+
 # If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
 adapter: lora
 # If you already have a lora model trained that you want to load, put that here.
@@ -659,7 +665,8 @@ wandb_mode: # "offline" to save run metadata locally and not sync to the server,
 wandb_project: # Your wandb project name
 wandb_entity: # A wandb Team name if using a Team
 wandb_watch:
-wandb_run_id: # Set the name of your wandb run
+wandb_name: # Set the name of your wandb run
+wandb_run_id: # Set the ID of your wandb run
 wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training
 
 # Where to save the full-finetuned model to
@@ -694,6 +701,9 @@ max_steps:
 eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
 eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
 
+loss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)
+loss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)
+
 # Save model as safetensors (require safetensors package)
 save_safetensors:
 
@@ -952,7 +962,7 @@ wandb_mode:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 ```
 

deepspeed/zero1.json

@@ -24,16 +24,6 @@
       "weight_decay": "auto"
     }
   },
-  "scheduler": {
-    "type": "WarmupDecayLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto",
-      "warmup_type": "linear",
-      "total_num_steps": "auto"
-    }
-  },
   "gradient_accumulation_steps": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",

deepspeed/zero2.json

@@ -28,16 +28,6 @@
       "weight_decay": "auto"
     }
   },
-  "scheduler": {
-    "type": "WarmupDecayLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto",
-      "warmup_type": "linear",
-      "total_num_steps": "auto"
-    }
-  },
   "gradient_accumulation_steps": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",

deepspeed/zero3.json

@@ -32,16 +32,6 @@
       "weight_decay": "auto"
     }
   },
-  "scheduler": {
-    "type": "WarmupDecayLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto",
-      "warmup_type": "linear",
-      "total_num_steps": "auto"
-    }
-  },
   "gradient_accumulation_steps": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",

examples/cerebras/btlm-ft.yml

@@ -35,7 +35,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 output_dir: btlm-out

examples/cerebras/qlora.yml

@@ -24,7 +24,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 batch_size: 4

examples/code-llama/13b/lora.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/code-llama/13b/qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/code-llama/34b/lora.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/code-llama/34b/qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/code-llama/7b/lora.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/code-llama/7b/qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/falcon/config-7b-lora.yml

@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2

examples/falcon/config-7b-qlora.yml

@@ -40,7 +40,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 

examples/falcon/config-7b.yml

@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2

examples/gptj/qlora.yml

@@ -21,7 +21,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 gradient_accumulation_steps: 2

examples/jeopardy-bot/config.yml

@@ -19,7 +19,7 @@ lora_fan_in_fan_out: false
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./jeopardy-bot-7b
 gradient_accumulation_steps: 1

examples/llama-2/fft_optimized.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 1

examples/llama-2/gptq-lora.yml

@@ -32,7 +32,7 @@ lora_target_linear:
 lora_fan_in_fan_out:
 wandb_project:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./model-out
 gradient_accumulation_steps: 1

examples/llama-2/lora.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/llama-2/qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/llama-2/relora.yml

@@ -35,7 +35,7 @@ relora_cpu_offload: false
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/llama-2/tiny-llama.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/mistral/config.yml

@@ -21,7 +21,7 @@ pad_to_sequence_len: true
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

examples/mistral/qlora.yml

@@ -38,7 +38,7 @@ lora_target_modules:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4
@@ -62,6 +62,9 @@ logging_steps: 1
 xformers_attention:
 flash_attention: true
 
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
 warmup_steps: 10
 eval_steps: 0.05
 eval_table_size:

examples/mpt-7b/config.yml

@@ -21,7 +21,7 @@ lora_fan_in_fan_out: false
 wandb_project: mpt-alpaca-7b
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./mpt-alpaca-7b
 gradient_accumulation_steps: 1

examples/openllama-3b/config.yml

@@ -23,7 +23,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./openllama-out
 gradient_accumulation_steps: 1

examples/openllama-3b/lora.yml

@@ -29,7 +29,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-out
 gradient_accumulation_steps: 1

examples/openllama-3b/qlora.yml

@@ -23,7 +23,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 gradient_accumulation_steps: 1

examples/phi/phi-ft.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 1

examples/phi/phi-qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 1

examples/pythia-12b/config.yml

@@ -24,7 +24,7 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./pythia-12b
 gradient_accumulation_steps: 1

examples/pythia/lora.yml

@@ -18,7 +18,7 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-alpaca-pythia
 gradient_accumulation_steps: 1

examples/qwen/lora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4
@@ -53,7 +53,7 @@ resume_from_checkpoint:
 local_rank:
 logging_steps: 1
 xformers_attention:
-flash_attention: true
+flash_attention:
 
 warmup_steps: 10
 eval_steps: 0.05

examples/qwen/qlora.yml

@@ -31,7 +31,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4
@@ -53,7 +53,7 @@ resume_from_checkpoint:
 local_rank:
 logging_steps: 1
 xformers_attention:
-flash_attention: true
+flash_attention:
 
 warmup_steps: 10
 eval_steps: 0.05

examples/redpajama/config-3b.yml

@@ -22,7 +22,7 @@ lora_fan_in_fan_out: false
 wandb_project: redpajama-alpaca-3b
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./redpajama-alpaca-3b
 batch_size: 4

examples/replit-3b/config-lora.yml

@@ -21,7 +21,7 @@ lora_fan_in_fan_out:
 wandb_project: lora-replit
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-replit
 batch_size: 8

examples/xgen-7b/xgen-7b-8k-qlora.yml

@@ -38,7 +38,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 

src/axolotl/cli/__init__.py

@@ -29,6 +29,7 @@ from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import is_main_process
 from axolotl.utils.models import load_tokenizer
 from axolotl.utils.tokenization import check_dataset_labels
+from axolotl.utils.trainer import prepare_optim_env
 from axolotl.utils.wandb_ import setup_wandb_env_vars
 
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
@@ -71,7 +72,7 @@ def do_merge_lora(
 
     LOG.info("running merge of LoRA with base model")
     model = model.merge_and_unload()
-    model.to(dtype=torch.float16)
+    model.to(dtype=cfg.torch_dtype)
 
     if cfg.local_rank == 0:
         LOG.info(f"saving merged model to: {str(Path(cfg.output_dir) / 'merged')}")
@@ -296,6 +297,8 @@ def load_cfg(config: Path = Path("examples/"), **kwargs):
 
     validate_config(cfg)
 
+    prepare_optim_env(cfg)
+
     normalize_config(cfg)
 
     setup_wandb_env_vars(cfg)

src/axolotl/core/trainer_builder.py

@@ -25,6 +25,7 @@ from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
 from axolotl.utils.callbacks import (
     EvalFirstStepCallback,
     GPUStatsCallback,
+    LossWatchDogCallback,
     SaveAxolotlConfigtoWandBCallback,
     SaveBetterTransformerModelCallback,
     bench_eval_callback_factory,
@@ -430,6 +431,9 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
             )
 
+        if self.cfg.loss_watchdog_threshold is not None:
+            callbacks.append(LossWatchDogCallback(self.cfg))
+
         return callbacks
 
     def get_post_trainer_create_callbacks(self, trainer):
@@ -643,7 +647,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         training_arguments_kwargs["group_by_length"] = self.cfg.group_by_length
         training_arguments_kwargs["report_to"] = "wandb" if self.cfg.use_wandb else None
         training_arguments_kwargs["run_name"] = (
-            self.cfg.wandb_run_id if self.cfg.use_wandb else None
+            self.cfg.wandb_name if self.cfg.use_wandb else None
         )
         training_arguments_kwargs["optim"] = (
             self.cfg.optimizer if self.cfg.optimizer else "adamw_hf"

src/axolotl/monkeypatch/cross_entropy.py

@@ -0,0 +1,168 @@
+# Adapted from Unsloth
+# https://github.com/unslothai/unsloth/blob/4b97a810b509c93f44be4c037c7aa18fb8922884/unsloth/kernels/cross_entropy_loss.py
+
+import triton
+import triton.language as tl
+import torch
+
+MAX_FUSED_SIZE = 65536
+
+def calculate_settings(n):
+    BLOCK_SIZE = triton.next_power_of_2(n)
+    # CUDA only supports 65536 - 2^16 threads per block
+    if BLOCK_SIZE > MAX_FUSED_SIZE:
+        raise RuntimeError(f"Cannot launch Triton kernel since n = {n} exceeds "\
+                           f"the maximum CUDA blocksize = {MAX_FUSED_SIZE}.")
+    num_warps = 4
+    if   BLOCK_SIZE >= 32768: num_warps = 32
+    elif BLOCK_SIZE >=  8192: num_warps = 16
+    elif BLOCK_SIZE >=  2048: num_warps = 8
+    return BLOCK_SIZE, num_warps
+pass
+
+@triton.jit
+def _cross_entropy_forward(logits_ptr, logits_row_stride,
+                           loss_ptr,
+                           lse_ptr,
+                           labels_ptr,
+                           n_cols,
+                           BLOCK_SIZE: tl.constexpr,):
+    """
+        Cross Entropy Loss = 1/n sum [ -yi log(Pi) ]
+        Pi = exp(xi) / sum(exp(xi))
+        CE_i = -y log(p) = -y log[ exp(x) / sum(exp(x)) ]
+             = -y [ x - log[sum(exp(x))] ]
+             = y * (log[sum(exp(x))] - x)
+        If y == 0: CE_i = 0
+        If y == 1: CE_i = logsumexp - x
+    """
+    row_idx = tl.program_id(0)
+    logits_ptr += row_idx * logits_row_stride
+    loss_ptr   += row_idx
+    lse_ptr    += row_idx
+    labels_ptr += row_idx
+
+    col_offsets = tl.arange(0, BLOCK_SIZE)
+    mask = col_offsets < n_cols
+
+    # TODO: Fixup int32 locations to int64
+    label_idx = tl.load(labels_ptr).to(tl.int32)
+    logits = tl.load(logits_ptr + col_offsets, mask = mask, other = -float("inf")).to(tl.float32)
+    max_logits = tl.max(logits, 0)
+    # Maximum stops overflow
+    lse = tl.log(tl.sum(tl.exp(logits - max_logits), 0)) + max_logits
+    tl.store(lse_ptr, lse)
+
+    if label_idx != -100:
+        logits_label = tl.load(logits_ptr + label_idx).to(tl.float32)
+        loss = lse - logits_label
+    else:
+        loss = 0.0
+    tl.store(loss_ptr, loss)
+pass
+
+
+@triton.jit
+def _cross_entropy_backward(logits_ptr, logits_row_stride,
+                            dloss_ptr,   dloss_row_stride,
+                            lse_ptr,
+                            labels_ptr,
+                            n_cols,
+                            BLOCK_SIZE: tl.constexpr,):
+    """
+        CE_i = -y log(P) = y * (log[sum(exp(x))] - x)
+        dC/dx = d/dx (y * log[sum(exp(x))] - x * y)
+
+        From https://en.wikipedia.org/wiki/LogSumExp
+        d/dx logsumexp = exp(x) / sum(exp(x)) = softmax(x)
+
+        dC/dx = y * exp(x) / sum(exp(x)) - d/dx (x * y)
+        dC/dx = y * exp[ log[exp(x) / sum(exp(x))] ] using x = exp(log(x)) trick
+        dC/dx = y * exp[x - logsumexp] - d/dx (x * y)
+
+        If y == 0: dC/dx = 0
+        If y == 1 and x == label: dC/dlabel = exp[x - logsumexp] - 1
+        If y == 1 and x != label: dC/dx     = exp[x - logsumexp]
+    """
+    row_idx = tl.program_id(0)
+    logits_ptr += row_idx * logits_row_stride
+    dloss_ptr  += row_idx *  dloss_row_stride
+    col_offsets = tl.arange(0, BLOCK_SIZE)
+    mask = col_offsets < n_cols
+    # TODO: Fixup int32 locations to int64
+    label_idx = tl.load(labels_ptr + row_idx).to(tl.int32)
+
+    if label_idx != -100:
+        dloss = tl.load(dloss_ptr)
+    else:
+        dloss = 0.0
+    logits = tl.load(logits_ptr + col_offsets, mask = mask, other = 0).to(tl.float32)
+    lse = tl.load(lse_ptr + row_idx)
+    probs = tl.exp(logits - lse)
+
+    probs = tl.where(col_offsets == label_idx, probs - 1.0, probs)
+    tl.store(logits_ptr + col_offsets, dloss * probs, mask = mask)
+
+
+
+class CrossEntropyLoss(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, logits, labels):
+        n_rows, n_cols = logits.shape
+        BLOCK_SIZE, num_warps = calculate_settings(n_cols)
+        losses    = torch.empty(n_rows, dtype = torch.float32, device = "cuda")
+        logsumexp = torch.empty(n_rows, dtype = torch.float32, device = "cuda")
+
+        _cross_entropy_forward[(n_rows,)](
+            logits, logits.stride(0),
+            losses,
+            logsumexp,
+            labels,
+            n_cols,
+            BLOCK_SIZE = BLOCK_SIZE,
+            num_warps  = num_warps,
+        )
+
+        ctx.BLOCK_SIZE = BLOCK_SIZE
+        ctx.num_warps = num_warps
+        ctx.save_for_backward(logits, logsumexp, labels)
+        return losses
+    pass
+
+    @staticmethod
+    def backward(ctx, dlosses):
+        logits, logsumexp, labels = ctx.saved_tensors
+        n_rows, n_cols = logits.shape
+
+        _cross_entropy_backward[(n_rows,)](
+            logits,   logits.stride(0),
+            dlosses, dlosses.stride(0),
+            logsumexp,
+            labels,
+            n_cols,
+            BLOCK_SIZE = ctx.BLOCK_SIZE,
+            num_warps  = ctx.num_warps,
+        )
+        return logits, None, None,
+    pass
+pass
+
+
+def fast_cross_entropy_loss(logits, labels):
+    """
+    Arguments:
+        logits: (batch, seq_len, vocab_size)
+        labels: (batch, seq_len,)
+    Returns:
+        losses: float
+    """
+    batch, seq_len, d = logits.shape
+    assert(labels.shape == (batch, seq_len))
+
+    loss = CrossEntropyLoss.apply(
+        logits.view(batch*seq_len, d),
+        labels.view(-1),
+    )
+    n_items = torch.count_nonzero(labels != -100)
+    return loss.sum() / n_items
+pass

src/axolotl/monkeypatch/mistral_attn_hijack_flash.py

@@ -13,16 +13,20 @@ from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-impor
     flash_attn_varlen_kvpacked_func,
     flash_attn_varlen_qkvpacked_func,
 )
-from transformers.modeling_outputs import BaseModelOutputWithPast
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
 from transformers.models.mistral.modeling_mistral import (
     MistralAttention as OriginalMistralAttention,
 )
 from transformers.models.mistral.modeling_mistral import (
     MistralDecoderLayer as OriginalMistralDecoderLayer,
 )
+from transformers.models.mistral.modeling_mistral import (
+    MistralForCausalLM as OriginalMistralForCausalLM,
+)
 from transformers.models.mistral.modeling_mistral import apply_rotary_pos_emb, repeat_kv
 
 from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+from axolotl.monkeypatch.cross_entropy import fast_cross_entropy_loss
 
 LOG = logging.getLogger("axolotl.monkeypatch.mistral")
 
@@ -36,6 +40,9 @@ def replace_mistral_attn_with_flash_attn(
     transformers.models.mistral.modeling_mistral.MistralAttention.forward = (
         flashattn_forward
     )
+    transformers.models.mistral.modeling_mistral.MistralForCausalLM.forward = (
+        mistral_causallm_forward
+    )
     if packed:
         transformers.models.mistral.modeling_mistral.MistralDecoderLayer = (
             MistralDecoderLayer
@@ -641,3 +648,71 @@ class MistralDecoderLayer(OriginalMistralDecoderLayer):
             outputs += (present_key_value,)
 
         return outputs
+
+def mistral_causallm_forward(
+    self: OriginalMistralForCausalLM,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+    *args, **kwargs
+) -> Union[Tuple, CausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+    ```"""
+
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+    )
+
+    hidden_states = outputs[0]
+    logits = self.lm_head(hidden_states)
+
+    loss = None
+    if labels is not None:
+        shift_logits = logits
+        if not hasattr(self, "extra_ignored_labels"):
+            self.extra_ignored_labels = torch.full((self.model.config.max_position_embeddings, 1), -100, device=shift_logits.device)
+        
+        shift_labels = torch.hstack((labels[..., 1:], self.extra_ignored_labels[:labels.shape[0]]))
+        shift_labels = shift_labels.to(shift_logits.device)
+
+        # FAST CROSS ENTROPY
+        loss = fast_cross_entropy_loss(shift_logits, shift_labels)
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        return (loss,) + output if loss is not None else output
+
+    return CausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )

src/axolotl/utils/callbacks.py

@@ -124,6 +124,36 @@ class GPUStatsCallback(
         return control
 
 
+class LossWatchDogCallback(TrainerCallback):
+    """Callback to track loss and stop training if loss is too high"""
+
+    def __init__(self, cfg):
+        self.cfg = cfg
+        self.logged = False
+        self.violations = 0
+        self.threshold = cfg.loss_watchdog_threshold
+        self.patience = cfg.loss_watchdog_patience or 3
+
+    def on_step_end(
+        self,
+        _args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **_kwargs,
+    ):
+        if len(state.log_history) > 0 and "loss" in state.log_history[-1]:
+            if state.log_history[-1]["loss"] > self.threshold:
+                self.violations += 1
+                if self.violations >= self.patience:
+                    LOG.warning(
+                        "Loss is too high, stopping training (loss_watchdog_threshold)"
+                    )
+                    control.should_training_stop = True
+            else:
+                self.violations = 0
+        return control
+
+
 def bench_eval_callback_factory(trainer, tokenizer):
     accuracy = evaluate.load("accuracy")
     abcd_idx = [

src/axolotl/utils/config.py

@@ -27,7 +27,7 @@ def choose_device(cfg):
 
     cfg.device = get_device()
     if cfg.world_size == 1:
-        cfg.device_map = "auto"
+        cfg.device_map = cfg.device_map or "auto"
     else:
         if cfg.device.startswith("cuda"):
             cfg.device_map = {"": torch.cuda.current_device()}
@@ -397,6 +397,13 @@ def validate_config(cfg):
             "Gradient checkpointing is broken for Qwen models for transformers>=4.35.0, except main branch."
         )
 
+    if cfg.wandb_run_id and not cfg.wandb_name:
+        cfg.wandb_name = cfg.wandb_run_id
+
+        LOG.warning(
+            "wandb_run_id sets the ID of the run. If you would like to set the name, please use wandb_name instead."
+        )
+
     # TODO
     # MPT 7b
     # https://github.com/facebookresearch/bitsandbytes/issues/25

src/axolotl/utils/data.py

@@ -698,24 +698,6 @@ def get_dataset_wrapper(
     return dataset_wrapper, dataset_prompter
 
 
-def encode_packed_pretraining(
-    tokenizer: PreTrainedTokenizerBase, max_tokens: int, examples: List[str]
-):
-    # tokenize all the examples
-    # rows get split with stride (overlap)
-    res = tokenizer(
-        examples,
-        truncation=True,
-        max_length=max_tokens,
-        add_special_tokens=True,
-        return_overflowing_tokens=True,
-        stride=256,
-    )
-    # convert to a dataset.from_list
-    # use a dataloader and multipack batch sampler to pack the data
-    pass
-
-
 def encode_pretraining(
     tokenizer: PreTrainedTokenizerBase, max_tokens: int, examples: List[str]
 ) -> Dict[str, List]:
@@ -831,7 +813,6 @@ def load_pretraining_dataset(path, tokenizer, max_tokens=2048, seed=42):
     dataset = dataset.map(
         encode,
         batched=True,
-        batch_size=10_000,
         input_columns="text",
         # remove all the existing columns after mapping since they end up having
         # a different length than the encoded/tokenized column

src/axolotl/utils/models.py

@@ -28,6 +28,27 @@ from axolotl.utils.dict import DictDefault
 LOG = logging.getLogger("axolotl")
 
 
+def check_model_config(cfg: DictDefault, model_config: AutoConfig):
+    quant_config_exists = hasattr(model_config, "quantization_config")
+    quant_config_method_is_gptq = (
+        quant_config_exists
+        and "quant_method" in model_config.quantization_config
+        and model_config.quantization_config["quant_method"] == "gptq"
+    )
+
+    if cfg.gptq and not quant_config_method_is_gptq:
+        raise ValueError(
+            "model_config.quantization_config is not set or quant_method is not set to gptq. "
+            "Please make sure to point to a GPTQ model."
+        )
+
+    if not cfg.gptq and quant_config_exists:
+        raise ValueError(
+            "model_config.quantization_config is set but `gptq` flag is not. "
+            "Please use the `gptq` flag to train quantized model or point to a non-quantized model."
+        )
+
+
 def load_model_config(cfg):
     model_config_name = cfg.base_model_config or cfg.base_model
     trust_remote_code = cfg.trust_remote_code is True
@@ -38,6 +59,8 @@ def load_model_config(cfg):
         for key, val in cfg.model_config.items():
             setattr(model_config, key, val)
 
+    check_model_config(cfg, model_config)
+
     return model_config
 
 
@@ -216,6 +239,7 @@ def load_model(
     model_kwargs = {}
 
     model_kwargs["device_map"] = cfg.device_map
+    model_kwargs["max_memory"] = cfg.max_memory
     model_kwargs["torch_dtype"] = cfg.torch_dtype
 
     if cfg.model_revision:
@@ -412,15 +436,22 @@ def load_model(
                 module.to(torch.float32)
 
     needs_fa2_dtype = cfg.adapter or cfg.fsdp
+    skip_prepare_model_for_kbit_training = False
+
+    if cfg.model_config_type == "qwen" and cfg.adapter == "lora":
+        # Qwen doesn't play nicely with LoRA if this is enabled
+        skip_prepare_model_for_kbit_training = True
+
     if (cfg.adapter == "lora" and load_in_8bit) or (
         cfg.adapter == "qlora" and cfg.load_in_4bit
     ):
         LOG.info("converting PEFT model w/ prepare_model_for_kbit_training")
         if cfg.gradient_checkpointing:
             model.gradient_checkpointing_enable()
-        model = prepare_model_for_kbit_training(
-            model, use_gradient_checkpointing=cfg.gradient_checkpointing
-        )
+        if not skip_prepare_model_for_kbit_training:
+            model = prepare_model_for_kbit_training(
+                model, use_gradient_checkpointing=cfg.gradient_checkpointing
+            )
         needs_fa2_dtype = True
 
     # LlamaRMSNorm layers are in fp32 after kbit_training or full finetune, so we need to

src/axolotl/utils/trainer.py

@@ -267,12 +267,14 @@ def setup_fsdp_envs(cfg):
         ] = cfg.fsdp_config.fsdp_transformer_layer_cls_to_wrap
 
 
-def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer, total_num_steps):
+def prepare_optim_env(cfg):
     if cfg.fsdp:
         setup_fsdp_envs(cfg)
     elif cfg.deepspeed:
         os.environ["ACCELERATE_USE_DEEPSPEED"] = "true"
 
+
+def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer, total_num_steps):
     trainer_builder = HFCausalTrainerBuilder(cfg, model, tokenizer)
     trainer_builder.train_dataset = train_dataset
     trainer_builder.eval_dataset = eval_dataset

src/axolotl/utils/wandb_.py

@@ -2,20 +2,20 @@
 
 import os
 
+from axolotl.utils.dict import DictDefault
 
-def setup_wandb_env_vars(cfg):
-    if cfg.wandb_mode and cfg.wandb_mode == "offline":
-        os.environ["WANDB_MODE"] = cfg.wandb_mode
-    elif cfg.wandb_project and len(cfg.wandb_project) > 0:
-        os.environ["WANDB_PROJECT"] = cfg.wandb_project
+
+def setup_wandb_env_vars(cfg: DictDefault):
+    for key in cfg.keys():
+        if key.startswith("wandb_"):
+            value = cfg.get(key, "")
+
+            if value and isinstance(value, str) and len(value) > 0:
+                os.environ[key.upper()] = value
+
+    # Enable wandb if project name is present
+    if cfg.wandb_project and len(cfg.wandb_project) > 0:
         cfg.use_wandb = True
-        if cfg.wandb_entity and len(cfg.wandb_entity) > 0:
-            os.environ["WANDB_ENTITY"] = cfg.wandb_entity
-        if cfg.wandb_watch and len(cfg.wandb_watch) > 0:
-            os.environ["WANDB_WATCH"] = cfg.wandb_watch
-        if cfg.wandb_log_model and len(cfg.wandb_log_model) > 0:
-            os.environ["WANDB_LOG_MODEL"] = cfg.wandb_log_model
-        if cfg.wandb_run_id and len(cfg.wandb_run_id) > 0:
-            os.environ["WANDB_RUN_ID"] = cfg.wandb_run_id
+        os.environ.pop("WANDB_DISABLED", None)  # Remove if present
     else:
         os.environ["WANDB_DISABLED"] = "true"

tests/test_validation.py

@@ -1,6 +1,7 @@
 """Module for testing the validation module"""
 
 import logging
+import os
 import unittest
 from typing import Optional
 
@@ -8,6 +9,7 @@ import pytest
 
 from axolotl.utils.config import validate_config
 from axolotl.utils.dict import DictDefault
+from axolotl.utils.wandb_ import setup_wandb_env_vars
 
 
 class ValidationTest(unittest.TestCase):
@@ -679,3 +681,83 @@ class ValidationTest(unittest.TestCase):
         )
 
         validate_config(cfg)
+
+
+class ValidationWandbTest(ValidationTest):
+    """
+    Validation test for wandb
+    """
+
+    def test_wandb_set_run_id_to_name(self):
+        cfg = DictDefault(
+            {
+                "wandb_run_id": "foo",
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert any(
+                "wandb_run_id sets the ID of the run. If you would like to set the name, please use wandb_name instead."
+                in record.message
+                for record in self._caplog.records
+            )
+
+            assert cfg.wandb_name == "foo" and cfg.wandb_run_id == "foo"
+
+        cfg = DictDefault(
+            {
+                "wandb_name": "foo",
+            }
+        )
+
+        validate_config(cfg)
+
+        assert cfg.wandb_name == "foo" and cfg.wandb_run_id is None
+
+    def test_wandb_sets_env(self):
+        cfg = DictDefault(
+            {
+                "wandb_project": "foo",
+                "wandb_name": "bar",
+                "wandb_run_id": "bat",
+                "wandb_entity": "baz",
+                "wandb_mode": "online",
+                "wandb_watch": "false",
+                "wandb_log_model": "checkpoint",
+            }
+        )
+
+        validate_config(cfg)
+
+        setup_wandb_env_vars(cfg)
+
+        assert os.environ.get("WANDB_PROJECT", "") == "foo"
+        assert os.environ.get("WANDB_NAME", "") == "bar"
+        assert os.environ.get("WANDB_RUN_ID", "") == "bat"
+        assert os.environ.get("WANDB_ENTITY", "") == "baz"
+        assert os.environ.get("WANDB_MODE", "") == "online"
+        assert os.environ.get("WANDB_WATCH", "") == "false"
+        assert os.environ.get("WANDB_LOG_MODEL", "") == "checkpoint"
+        assert os.environ.get("WANDB_DISABLED", "") != "true"
+
+    def test_wandb_set_disabled(self):
+        cfg = DictDefault({})
+
+        validate_config(cfg)
+
+        setup_wandb_env_vars(cfg)
+
+        assert os.environ.get("WANDB_DISABLED", "") == "true"
+
+        cfg = DictDefault(
+            {
+                "wandb_project": "foo",
+            }
+        )
+
+        validate_config(cfg)
+
+        setup_wandb_env_vars(cfg)
+
+        assert os.environ.get("WANDB_DISABLED", "") != "true"