dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[fix] t2_n_2 max_num_seqs; t13

Browse Source
This commit is contained in:
root
2026-08-24 04:18:10 +00:00
parent 3fe0590266
commit 367031b3bc
5 changed files with 884 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
qwen3_6_scripts/patch_ops.sh
View File

@@ -162,6 +162,12 @@ python3 ./patch_worker_cache_transfer_order.py
# when context length is high
cp ./paged_attn.py "${VLLM_ROOT}/attention/ops/paged_attn.py"
# --- multimodal: increase image fetch timeout and add retry ------------------
# Container networks often cannot reach remote image URLs within the default
# 5-second timeout. Raise to 30s and retry 3 times.
cp ./multimodal_utils.py "${VLLM_ROOT}/multimodal/utils.py"
cp ./envs.py "${VLLM_ROOT}/envs.py"
# --- model_runner.py: fix prefix_cache_hit stays True in chunked-prefill chunk 2+ ---
# Bug: _compute_for_prefix_cache_hit Case 1 (prefix_cache_len <= context_len)
# leaves prefix_cache_hit=True. Then _add_seq_group uses block_table=computed_block_nums