mirror of
https://github.com/deepseek-ai/DeepInfer.git
synced 2026-10-30 15:57:20 +08:00
The number of RDMA QPs DeepEP allocates for the Engram buffer is derived from its max_num_tokens, which in turn comes from max_num_batched_tokens (model_loader.py). Raising the QP count therefore required raising the global batched-token budget, which also inflates the EP/MoE/CP buffers. Add DEEPINFER_ENGRAM_MIN_PREFILLING_TOKENS as a lower bound applied only to the Engram buffer, so a deployment can give Engram more QPs without resizing the other buffers. Default 0 keeps the current behavior. |
||
|---|---|---|
| .. | ||
| _C | ||
| distributed | ||
| engine | ||
| entrypoints | ||
| hugging_face | ||
| model_executor | ||
| npu_ops | ||
| structure_output | ||
| ve | ||
| worker | ||
| __init__.py | ||
| attrdict.py | ||
| config.py | ||
| environ.py | ||
| load_class.py | ||
| logger.py | ||
| model_config_loader.py | ||
| multimodal_tokens.py | ||
| platform.py | ||
| py.typed | ||
| sampling_params.py | ||
| sequence.py | ||
| shared_memory.py | ||
| testing.py | ||
| timer.py | ||
| utils.py | ||