fix: change DeepSeek default aux model from deepseek-chat to deepseek-v4-flash - #54556
fix: change DeepSeek default aux model from deepseek-chat to deepseek-v4-flash#54556ycj wants to merge 1 commit into
Conversation
…-v4-flash deepseek-v4-flash is a modern V4-generation model that offers better capability than V3 (deepseek-chat) while still being faster and cheaper than V4 Pro. This makes it a better default for auxiliary tasks like context compression and vision fallback. - Update default_aux_model in DeepSeekProfile - Update corresponding test assertions
There was a problem hiding this comment.
Pull request overview
Updates the DeepSeek model-provider profile so Hermes’ auxiliary-task router (context compression, vision fallback, etc.) defaults to a newer V4-generation DeepSeek model instead of the legacy V3 model.
Changes:
- Switch
deepseekprovider profiledefault_aux_modelfromdeepseek-chattodeepseek-v4-flash. - Update DeepSeek provider-profile tests to assert the new default aux model value.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| plugins/model-providers/deepseek/init.py | Changes DeepSeek default_aux_model to deepseek-v4-flash. |
| tests/plugins/model_providers/test_deepseek_profile.py | Updates assertions (and test names) to match the new DeepSeek aux-model default. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
alt-glitch
added
type/feature
Related: #49169 (open) already changes |
tonydwb
left a comment
tonydwb
left a comment
There was a problem hiding this comment.
Code Review Summary
Verdict: LGTM
Simple, clean change — updates the default DeepSeek aux model from deepseek-chat to deepseek-v4-flash. Tests are updated to match. No security concerns.
Note: prior COMMENT review exists on this PR. This is a new formal review.
Reviewed by Hermes Agent
teknium1
left a comment
teknium1
left a comment
There was a problem hiding this comment.
Thanks for the focused provider update. The changed profile field is the active auxiliary-model source: agent/auxiliary_client.py:448-455 resolves the registered profile's default_aux_model.
Problems
tests/plugins/model_providers/test_deepseek_profile.py:199and:203freeze a specific model identifier. These are change-detector assertions, contrary to the behavior-contract guidance inAGENTS.md:80-83andAGENTS.md:1309-1356.
Suggested changes
- Assert that
_get_aux_model_for_provider("deepseek")equalsdeepseek_profile.default_aux_model, and that the profile default belongs to the curated DeepSeek catalog inhermes_cli/models.py:380-385, rather than pinning the model string twice.
Automated hermes-sweeper review.
| def test_profile_advertises_deepseek_chat(self, deepseek_profile): | ||
| assert deepseek_profile.default_aux_model == "deepseek-chat" | ||
| def test_profile_advertises_deepseek_v4_flash(self, deepseek_profile): | ||
| assert deepseek_profile.default_aux_model == "deepseek-v4-flash" |
There was a problem hiding this comment.
This fixed model-name assertion is a change-detector snapshot. Please test the resolver/profile relationship instead (and, if useful, that the profile default is in the DeepSeek curated catalog) so routine model-default updates do not require rewriting this test.
teknium1
added
sweeper:risk-compatibility
Summary
Change the DeepSeek provider's
default_aux_modelfromdeepseek-chat(V3) todeepseek-v4-flash(V4).Motivation
deepseek-v4-flashis a modern V4-generation model that offers:deepseek-chat)This makes it a better default for auxiliary tasks like context compression, vision fallback, and other side-jobs that use the provider's default model.
Changes
plugins/model-providers/deepseek/__init__.py:default_aux_model→deepseek-v4-flashtests/plugins/model_providers/test_deepseek_profile.py: update two test assertions to matchTest Plan
→ 29 tests passed, 0 failed