mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-07-15 14:22:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

test: fold deepseek-v4 cases into canonical reasoning-floor test, drop duplicate file

Browse source 
The salvaged PR added a standalone test_reasoning_timeouts.py that duplicated
the structure of the existing parametrized test_reasoning_stale_timeout_floor.py.
Fold the v4-flash/v4-pro/-free positive cases and deepseek-chat negative cases
into the canonical parametrized tables and remove the redundant file.
This commit is contained in:
kshitijk4poor 2026-07-09 14:58:19 +05:30 • committed by kshitij
parent 1e16120603
commit 411d599764
2 changed files with 10 additions and 47 deletions
Download patch file Download diff file Expand all files Collapse all files

11
tests/agent/test_reasoning_stale_timeout_floor.py
View file

@ -46,10 +46,15 @@ import pytest
("nvidia/nemotron-3-ultra-550b-a55b", 600.0),
("nvidia/nemotron-3-super-120b-a12b", 600.0),
("nvidia/nemotron-3-nano-30b-a3b", 300.0),
# DeepSeek R1 + DeepSeek reasoner.
# DeepSeek R1 + DeepSeek reasoner + V4 reasoning series.
# V4 emits reasoning_content in a separate delta before final
# content (same shape as R1), so it needs the same 600s floor.
("deepseek/deepseek-r1", 600.0),
("deepseek/deepseek-r1-distill-llama-70b", 600.0),
("deepseek/deepseek-reasoner", 600.0),
("deepseek/deepseek-v4-flash", 600.0),
("deepseek/deepseek-v4-pro", 600.0),
("deepseek-v4-flash-free", 600.0), # catalog -free variant inherits via separator anchor
# Qwen QwQ + Qwen3 thinking variants (qwen3 family entry matches all).
("qwen/qwq-32b-preview", 300.0),
("qwen/qwen3-235b-a22b-thinking", 180.0),
@ -104,6 +109,10 @@ def test_reasoning_stale_timeout_floor_positive_cases(model, expected):
"x-ai/grok-code-fast-1",
# Qwen2 must not match Qwen3 (different family).
"qwen2-72b-instruct",
# Non-reasoning DeepSeek chat must not match the v4 reasoning entries.
"deepseek-chat",
"deepseek/deepseek-chat",
"some-deepseek-v4-flash", # embedded v4 slug, NOT start of slug
# Empty / None / non-string inputs — must return None, not raise.
"",
None,

46
tests/agent/test_reasoning_timeouts.py
View file

@ -1,46 +0,0 @@
"""Test get_reasoning_stale_timeout_floor for DeepSeek V4 reasoning models."""
from agent.reasoning_timeouts import get_reasoning_stale_timeout_floor
def test_deepseek_v4_reasoning_models_have_timeout_floor():
"""DeepSeek V4 reasoning models should have 600s timeout floor.
DeepSeek V4 models (deepseek-v4-flash, deepseek-v4-pro) emit
reasoning_content in a separate delta field before final content,
requiring the same extended stale timeout floor as R1.
See #60338.
"""
# Direct model names (no aggregator prefix)
assert get_reasoning_stale_timeout_floor("deepseek-v4-flash") == 600.0
assert get_reasoning_stale_timeout_floor("deepseek-v4-pro") == 600.0
# With aggregator prefixes (common usage)
assert get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-flash") == 600.0
assert get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-pro") == 600.0
# With custom provider prefixes
assert get_reasoning_stale_timeout_floor("opencode/deepseek-v4-flash") == 600.0
assert get_reasoning_stale_timeout_floor("custom/deepseek-v4-pro") == 600.0
def test_deepseek_v4_timeout_floor_matches_r1():
"""V4 models should have the same floor as R1 (600s)."""
v4_flash_floor = get_reasoning_stale_timeout_floor("deepseek-v4-flash")
v4_pro_floor = get_reasoning_stale_timeout_floor("deepseek-v4-pro")
r1_floor = get_reasoning_stale_timeout_floor("deepseek-r1")
reasoner_floor = get_reasoning_stale_timeout_floor("deepseek-reasoner")
assert v4_flash_floor == v4_pro_floor == r1_floor == reasoner_floor == 600.0
def test_deepseek_v4_does_not_match_non_reasoning_deepseek():
"""Non-reasoning deepseek variants should not match V4 patterns."""
# deepseek-chat is not a reasoning model and should not match
# the deepseek-v4 patterns (requires start-of-slug anchor).
assert get_reasoning_stale_timeout_floor("deepseek-chat") is None
assert get_reasoning_stale_timeout_floor("deepseek/deepseek-chat") is None
# But V4 patterns should still match V4 models even with non-V4 prefixes
assert get_reasoning_stale_timeout_floor("some-deepseek-v4-flash") is None # wrong prefix
assert get_reasoning_stale_timeout_floor("deepseek-v4-flash-model") == 600.0 # suffix OK