EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[Fixbug] Fix accuracy for DeepSeek-V2-Lite (#3016)

Browse Source 
### What this PR does / why we need it?

Fix accuracy for DeepSeek-V2-Lite

### Does this PR introduce _any_ user-facing change?
no

### How was this patch tested?
CI passed

- vLLM version: v0.10.2
- vLLM main:
66072b36db

Signed-off-by: hfadzxy <starmoon_zhang@163.com>
This commit is contained in:
zhangxinyuehfad
2025-09-18 23:58:23 +08:00
committed by GitHub
parent 367edff5af
commit 0c04bf1e36
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/e2e/models/configs/DeepSeek-V2-Lite.yaml
View File

@@ -7,6 +7,8 @@ tasks:
- name: "exact_match,flexible-extract" - name: "exact_match,flexible-extract"
value: 0.375 value: 0.375
tensor_parallel_size: 2 tensor_parallel_size: 2
batch_size: 8
gpu_memory_utilization: 0.7
apply_chat_template: False apply_chat_template: False
fewshot_as_multiturn: False fewshot_as_multiturn: False
trust_remote_code: True trust_remote_code: True

4
tests/e2e/models/test_lm_eval_correctness.py
View File

@@ -84,7 +84,7 @@ def generate_report(tp_size, eval_config, report_data, report_dir, env_config):
apply_chat_template=eval_config.get("apply_chat_template", True), apply_chat_template=eval_config.get("apply_chat_template", True),
fewshot_as_multiturn=eval_config.get("fewshot_as_multiturn", True), fewshot_as_multiturn=eval_config.get("fewshot_as_multiturn", True),
limit=eval_config.get("limit", "N/A"), limit=eval_config.get("limit", "N/A"),
batch_size="auto", batch_size=eval_config.get("batch_size", "auto"),
num_fewshot=eval_config.get("num_fewshot", "N/A"), num_fewshot=eval_config.get("num_fewshot", "N/A"),
rows=report_data["rows"], rows=report_data["rows"],
parallel_mode=parallel_mode) parallel_mode=parallel_mode)
@@ -110,7 +110,7 @@ def test_lm_eval_correctness_param(config_filename, tp_size, report_dir,
"apply_chat_template": eval_config.get("apply_chat_template", True), "apply_chat_template": eval_config.get("apply_chat_template", True),
"fewshot_as_multiturn": eval_config.get("fewshot_as_multiturn", True), "fewshot_as_multiturn": eval_config.get("fewshot_as_multiturn", True),
"limit": eval_config.get("limit", None), "limit": eval_config.get("limit", None),
"batch_size": "auto", "batch_size": eval_config.get("batch_size", "auto"),
} }
for s in ["num_fewshot", "fewshot_as_multiturn", "apply_chat_template"]: for s in ["num_fewshot", "fewshot_as_multiturn", "apply_chat_template"]:
val = eval_config.get(s, None) val = eval_config.get(s, None)