MiniCPM5-2B: spec + validated attn tuple (128, 16, 8, 128, qk_norm=False) - #85
Merged
Merged
Conversation
Atomic-Germ
reviewed
Sep 14, 2026
Collaborator
|
yeah I didn't ever do anything with it, so this PR is very welcome! It's a good model. |
Cyronius
approved these changes
Sep 15, 2026
Cyronius
pushed a commit
to Cyronius/OpenFlowLM-Next
that referenced
this pull request
Sep 16, 2026
main added MiniCPM5-2B's attention tuple to the catalogue (Atomic-Germ#85) while this branch widened every tuple in that list by one field, so the two landed on the same lines. MiniCPM5's tuple carries over as-is with qkv_bias False, in the position main gave it. The new recipe spec that came with it, minicpm5-2b.json, wants a line in test_no_shipped_models_hash_moved's frozen list. The eleven hashes that predate this branch are unchanged, which is the answer to what that test asks: qkv_bias is a kernel-template knob, not a ModelSpec field, so it moved none of them. 602 pass. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #83.
Adds the MiniCPM5-2B spec plus the attention tuple it needs, measured end-to-end on hardware (Strix Halo, XDNA2, fw 1.1.2.65, in-tree amdxdna 0.7, XRT 2.26):
open_kernels/export_qwen36_kernels.py --spec open_kernels/recipes/specs/minicpm5-2b.json→ 42/42 layers resident4· "Was ist 25*14?" →350· coherent German/English reasoning; CPU reference (llama.cpp, same GGUF) agreesThe first kernel build used
OPEN_KERNELS_UNVALIDATED=1per the documented override; this PR adds the catalogue point so future builds validate normally.catalogue.py: adds(128, 16, 8, 128, False, False, False)next to the Qwen3-1.7B point (same geometry,qk_normoff).Happy to run any additional fixture/fp64 comparison you'd like.