Commit History
3 days ago
-
Revert the merge of discussion #1
Verified signatureRestores README.md, d3q/load.py and MODEL_MANIFEST.json as of b5aebab7f4; the discussion was merged by mistake.8acfecc -
Card: first-start note, same-machine vision baseline, Typed Decisions note; drop unmeasured throughput line (#1)
- Card: first-start note, same-machine vision baseline, Typed Decisions note; drop unmeasured throughput line (98b3e44d8ab4ab05f37430cc1672769b16b44f6d) - Package: drop internal names and stale files; MODEL_MANIFEST.json entries for the changed files (466bad92b791ae516e51481dabc6116849e2e027) Co-authored-by: Hsiung Li Syun <TheoHsiung@users.noreply.huggingface.co>9497574
4 days ago
-
Add Typed Decisions eval results
Verified signatureLocalLLaMA/typed-decisions test split (400 cases, 2,000 decisions), scored with the benchmark run's score_model.py against this package served by the vLLM Semantic Router model runtime on one AMD Instinct MI300X. d3 on the same setup scores accuracy 0.7485, KL 0.4096, Brier 0.1457, ECE 0.0510, against 0.749 / 0.4093 / 0.1456 / 0.0486 in the official vLLM run.b5aebab -
Card: link the merged record on vllm-project/semantic-router main
Verified signature9a83ee5
6 days ago
-
Label the d3 column of the test split table as BF16
Verified signature2fb271a -
Add the Perception Test validation results
Verified signature6f4a524
7 days ago
-
Measure serving latency and memory on MI300X and RTX PRO 6000 as the GPTQ card does
Verified signature781a56f -
Index FP8 activation rows in 64 bits and add Decision Index results
Verified signature5887468
8 days ago
-
Runtime: loading checks only the files the runtime reads
Verified signaturefb2f5e8 -
Card: serve with vllm-sr serve -e
Verified signature54be3ed -
Card: the revision to serve
Verified signatureab1abc8 -
Upload d3-FP8
Verified signature94c520d -
initial commit
Verified signaturec63b222