also upgrade accelerate

Co-authored-by: NanoCode012 <nano@axolotl.ai>

README.md

@@ -121,7 +121,7 @@ Features:
 
 Get started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.
 
-**Requirements**: Python >=3.10 and Pytorch >=2.1.1.
+**Requirements**: Nvidia GPU (Ampere architecture or newer for `bf16` and Flash Attention), Python >=3.10 and PyTorch >=2.3.1.
 
 ```bash
 git clone https://github.com/axolotl-ai-cloud/axolotl
@@ -383,7 +383,7 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
         - typescript
       type: ... # unimplemented custom format
 
-      # fastchat conversation (deprecation soon, use chat_template)
+      # fastchat conversation (deprecation soon, use chat_template https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/conversation.html#chat_template)
       # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
     - path: ...
       type: sharegpt

examples/deepseek-v2/qlora-fsdp-2_5.yaml

@@ -9,7 +9,7 @@ strict: false
 plugins:
   - axolotl.integrations.liger.LigerPlugin
 liger_rms_norm: true
-liger_swiglu: true
+liger_glu_activation: true
 liger_fused_linear_cross_entropy: true
 
 chat_template: deepseek_v2

examples/llama-3/fft-8b-liger-fsdp.yaml

@@ -9,21 +9,27 @@ liger_fused_linear_cross_entropy: true
 
 strict: false
 
+chat_template: llama3
 datasets:
-    - path: tatsu-lab/alpaca
-      type: alpaca
+  - path: mlabonne/FineTome-100k
+    type: chat_template
+    split: train[:20%]
+    field_messages: conversations
+    message_field_role: from
+    message_field_content: value
+
 dataset_prepared_path: last_run_prepared
-val_set_size: 0
+val_set_size: 0.02
 output_dir: ./outputs/out
 
 sequence_len: 4096
 sample_packing: true
 pad_to_sequence_len: true
 
-wandb_project: check_liger_hf_GA_llama_fix-3
-wandb_entity: axolotl-ai
+wandb_project:
+wandb_entity:
 wandb_watch:
-wandb_name: pr/fix333-tr4.46.1
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4

requirements.txt

@@ -1,10 +1,10 @@
 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
 packaging==23.2
 peft==0.13.2
-transformers==4.46.1
+transformers==4.46.2
 tokenizers>=0.20.1
 bitsandbytes==0.44.1
-accelerate==1.0.1
+accelerate==1.1.0
 datasets==3.0.1
 deepspeed==0.15.3
 pydantic==2.6.3
@@ -33,8 +33,8 @@ gradio==3.50.2
 tensorboard
 python-dotenv==1.0.1
 autoawq>=0.2.5
-triton>=3.1.0
-liger-kernel==0.3.1
+triton>=2.3.0
+liger-kernel==0.4.0
 
 mamba-ssm==1.2.0.post1
 

src/axolotl/core/trainer_builder.py

@@ -896,13 +896,13 @@ class AxolotlTrainer(SchedulerMixin, Trainer):
         for key, value in metrics.items():
             self._stored_metrics[train_eval][key].append(value)
 
-    def _save_checkpoint(self, model, trial, metrics=None):
+    def _save_checkpoint(self, model, trial):
         # make sure the checkpoint dir exists, since trainer is flakey
         checkpoint_folder = f"{PREFIX_CHECKPOINT_DIR}-{self.state.global_step}"
         run_dir = self._get_output_dir(trial=trial)
         output_dir = os.path.join(run_dir, checkpoint_folder)
         os.makedirs(output_dir, exist_ok=True)
-        return super()._save_checkpoint(model, trial, metrics=metrics)
+        return super()._save_checkpoint(model, trial)
 
 
 class AxolotlMambaTrainer(AxolotlTrainer):

tests/e2e/integrations/liger.py

@@ -1,7 +1,6 @@
 """
 Simple end-to-end test for Liger integration
 """
-
 import unittest
 from pathlib import Path
 

tests/integrations/liger.py

@@ -0,0 +1,80 @@
+"""
+config validation tests for swiglu args
+"""
+# pylint: disable=duplicate-code
+import logging
+from typing import Optional
+
+import pytest
+
+from axolotl.utils.config import validate_config
+from axolotl.utils.dict import DictDefault
+
+
+@pytest.fixture(name="minimal_base_cfg")
+def fixture_cfg():
+    return DictDefault(
+        {
+            "base_model": "TinyLlama/TinyLlama-1.1B-Chat-v0.6",
+            "learning_rate": 0.000001,
+            "datasets": [
+                {
+                    "path": "mhenrichsen/alpaca_2k_test",
+                    "type": "alpaca",
+                }
+            ],
+            "micro_batch_size": 1,
+            "gradient_accumulation_steps": 1,
+        }
+    )
+
+
+class BaseValidation:
+    """
+    Base validation module to setup the log capture
+    """
+
+    _caplog: Optional[pytest.LogCaptureFixture] = None
+
+    @pytest.fixture(autouse=True)
+    def inject_fixtures(self, caplog):
+        self._caplog = caplog
+
+
+# pylint: disable=too-many-public-methods
+class TestValidation(BaseValidation):
+    """
+    Test the validation module for liger
+    """
+
+    def test_deprecated_swiglu(self, minimal_cfg):
+        test_cfg = DictDefault(
+            {
+                "liger_swiglu": False,
+            }
+            | minimal_cfg
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            updated_cfg = validate_config(test_cfg)
+            assert (
+                "The 'liger_swiglu' argument is deprecated"
+                in self._caplog.records[0].message
+            )
+            assert updated_cfg.liger_swiglu is None
+            assert updated_cfg.liger_glu_activations is False
+
+    def test_conflict_swiglu_ligergluactivation(self, minimal_cfg):
+        test_cfg = DictDefault(
+            {
+                "liger_swiglu": False,
+                "liger_glu_activations": True,
+            }
+            | minimal_cfg
+        )
+
+        with pytest.raises(
+            ValueError,
+            match=r".*You cannot have both `liger_swiglu` and `liger_glu_activation` set.*",
+        ):
+            validate_config(test_cfg)