Config: evalscope generation_config format + our extras (repeats, env) at bench level; parallel_tool_calls noted as unsupported in adapter
Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
parent
0ada7d8a0e
commit
0815f6e4f8
@ -16,12 +16,14 @@ aime24:
|
|||||||
top_p: 1.0
|
top_p: 1.0
|
||||||
stream: true
|
stream: true
|
||||||
max_tokens: 32768
|
max_tokens: 32768
|
||||||
|
repeats: 12
|
||||||
aime25:
|
aime25:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 1.0
|
temperature: 1.0
|
||||||
top_p: 1.0
|
top_p: 1.0
|
||||||
stream: true
|
stream: true
|
||||||
max_tokens: 32768
|
max_tokens: 32768
|
||||||
|
repeats: 12
|
||||||
mmlu_pro:
|
mmlu_pro:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 0.0
|
temperature: 0.0
|
||||||
@ -61,12 +63,14 @@ aime26:
|
|||||||
top_p: 1.0
|
top_p: 1.0
|
||||||
stream: true
|
stream: true
|
||||||
max_tokens: 32768
|
max_tokens: 32768
|
||||||
|
repeats: 12
|
||||||
hmmt26:
|
hmmt26:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 1.0
|
temperature: 1.0
|
||||||
top_p: 1.0
|
top_p: 1.0
|
||||||
stream: true
|
stream: true
|
||||||
max_tokens: 32768
|
max_tokens: 32768
|
||||||
|
repeats: 12
|
||||||
imo_answerbench:
|
imo_answerbench:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 1.0
|
temperature: 1.0
|
||||||
@ -164,10 +168,8 @@ tau2_bench:
|
|||||||
generation_config:
|
generation_config:
|
||||||
temperature: 0.0
|
temperature: 0.0
|
||||||
max_tokens: 4096
|
max_tokens: 4096
|
||||||
agent_config:
|
env: tau2_official
|
||||||
mode: native
|
max_turns: 50
|
||||||
strategy: react
|
|
||||||
max_steps: 50
|
|
||||||
general_fc:
|
general_fc:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 0.0
|
temperature: 0.0
|
||||||
@ -185,6 +187,7 @@ bfcl_v3:
|
|||||||
extra_params:
|
extra_params:
|
||||||
is_fc_model: true
|
is_fc_model: true
|
||||||
underscore_to_dot: true
|
underscore_to_dot: true
|
||||||
|
env: bfcl_mock
|
||||||
swe_bench_verified:
|
swe_bench_verified:
|
||||||
generation_config:
|
generation_config:
|
||||||
temperature: 0.0
|
temperature: 0.0
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user