NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-19 12:57:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[pre-commit.ci] auto fixes from pre-commit.com hooks

Browse source 
for more information, see https://pre-commit.ci
This commit is contained in:
pre-commit-ci[bot] 2026-01-19 19:25:17 +00:00 • committed by Jai Suphavadeeprasit
parent f87455a712
commit e6e0691bd7
2 changed files with 18 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

10
example_trainer/README.md
View file

@ -816,7 +816,7 @@ pkill -9 -u $USER -f "vllm|grpo|python|run-api"
## Feature Availability Matrix
### What's Available
### What's Available
| Feature | Status | Notes |
|---------|--------|-------|
@ -832,7 +832,7 @@ pkill -9 -u $USER -f "vllm|grpo|python|run-api"
| **Wandb Logging** | Working | Via `--use-wandb` flag |
| **Custom Environments** | Working | Extend `BaseEnv` class |
### What's NOT Available
### What's NOT Available
| Feature | Mode | Status | Reason / Workaround |
|---------|------|--------|---------------------|
@ -858,7 +858,7 @@ pkill -9 -u $USER -f "vllm|grpo|python|run-api"
| **LoRA** | Supported | Via vLLM | Multiple Trainers |
| **Legacy** | Supported | Via vLLM | Multiple Trainers |
> **Key Point**: The multi-GPU limitation is **ONLY for single-copy mode** due to CUDA IPC constraints.
> **Key Point**: The multi-GPU limitation is **ONLY for single-copy mode** due to CUDA IPC constraints.
> LoRA and Legacy modes work with standard vLLM which fully supports tensor parallelism.
#### Pipeline Parallel (PP)
@ -956,7 +956,7 @@ CUDA_VISIBLE_DEVICES=5 python -u example_trainer/grpo.py \
## Future Work
### High Priority
### High Priority
| Feature | Description |
|---------|-------------|
@ -964,7 +964,7 @@ CUDA_VISIBLE_DEVICES=5 python -u example_trainer/grpo.py \
| **Automatic Server Type Detection** | Auto-detect correct `server_type` for environments |
| **Checkpoint Resume** | Resume training from checkpoints seamlessly |
### Medium Priority
### Medium Priority
| Feature | Description | Difficulty |
|---------|-------------|------------|

20
example_trainer/grpo.py
View file

@ -159,7 +159,7 @@ class TrainingConfig(BaseModel):
"and contains CUDA IPC handles for single-copy mode."
),
)
# Debug flags
debug_loading: bool = Field(
False,
@ -1309,7 +1309,9 @@ def finalize_training(
print(f"\n{'='*70}")
print(f"BENCHMARK SUMMARY ({mode})")
print(f"{'='*70}")
print(f" Total training time: {total_time:.2f}s ({total_time/60:.2f} min)")
print(
f" Total training time: {total_time:.2f}s ({total_time/60:.2f} min)"
)
print(f" Total steps: {total_steps}")
print(f" ")
print(f" TIMING BREAKDOWN:")
@ -1478,7 +1480,11 @@ def train(config: TrainingConfig):
model, tokenizer, config.save_path, config.training_steps, is_final=True
)
finalize_training(
use_wandb, training_start_time, "legacy", config.training_steps, benchmark_stats,
use_wandb,
training_start_time,
"legacy",
config.training_steps,
benchmark_stats,
benchmark=config.benchmark,
)
@ -2161,10 +2167,10 @@ def config_from_args(args: argparse.Namespace) -> TrainingConfig:
lora_alpha=args.lora_alpha,
lora_dropout=args.lora_dropout,
lora_target_modules=args.lora_target_modules,
single_copy=getattr(args, 'single_copy', False),
vllm_config_path=getattr(args, 'vllm_config_path', None),
debug_loading=getattr(args, 'debug_loading', False),
benchmark=getattr(args, 'benchmark', False),
single_copy=getattr(args, "single_copy", False),
vllm_config_path=getattr(args, "vllm_config_path", None),
debug_loading=getattr(args, "debug_loading", False),
benchmark=getattr(args, "benchmark", False),
)