Jev directory
CommunityJev Models & Reimplementations
Jev Boost to8 abtrain apibank Qwen3.6-35B-A3B-FP8 Embeddings
Hugging Face dataset of 6,919 API-Bank training decisions rendered in the Jev Decision Index shape and embedded with Qwen/Qwen3.6-35B-A3B-FP8 at 2048 dimensions, for post-training CLM heads.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- 6,919 train and 809 validation decision pairs, each with up to 8 hard negatives (negatives_k 8).
- Vectors are 2048-d float16, L2-normalized, pooled from Qwen/Qwen3.6-35B-A3B-FP8 with vLLM.
- Built from API-Bank's lv1-3 api-train dialogues; the benchmark's test data is not read.
- One-token actions and negatives are zeroed and resolved from a separate single-token embedding dataset on load.
- Total file size 321 MB; license follows the source, MIT (API-Bank, Li et al. 2023).
Questions about this entry
Short answers drawn from the linked project or page.
A Hugging Face dataset from clm-project holding API-Bank training decisions rendered in the Jev Decision Index shape and embedded with Qwen/Qwen3.6-35B-A3B-FP8. It is intended as post-training data for CLM heads through clm_train.data.jev_boost.
The card gives the command clm-train posttrain --source hf:boost-to8-abtrain-apibank-qwen3.6. Files are split into state_embeddings.pt, action_embeddings.pt, neg_embeddings.pt, neg_counts.pt, single_tokens.npz and metadata.json, with train and validation splits.
Qwen/Qwen3.6-35B-A3B-FP8, embedded with vLLM pooling using the last token, L2-normalized to 2048 dimensions and stored as float16. The card lists hidden_size 2048, max_model_len 16,384 and state_head_tokens 2,048.
API-Bank (liminghao1630/API-Bank, revision 12e8158b7628c168f07e8f31fbbe3445e99f44cf), split train, using its lv1-3 api-train dialogues. The benchmark's test data is not read. Each decision merges the dialogue's own tools into the index's 53-tool catalog.
One-token sequences are never embedded, because vLLM pooling of a one-token prompt is noise for this model. Those rows are zeroed and single_tokens.npz points them at jackyk02/qwen3.6-35b-a3b-fp8-single-token-embeddings, which clm_train.data.stores.EmbeddingDir applies on load.
The source's license, MIT (API-Bank, Li et al. 2023). The vectors are a derivative of that text, so the same terms carry over to the embeddings.
At most 20,000 rows are kept via a hash sample, and each row carries at most 8 hard negatives, zero-padded in neg_embeddings.pt. Near-copies of the gold option are dropped, and items whose state shares a paragraph with a Jev suite item are dropped.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A Hugging Face dataset of Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings built from AirDialogue's book, change, and cancel requests, shaped as When2Call multiple-choice decisions for post-training CLM heads.
Jev Models & Reimplementations#huggingface#open-models#benchmark
Community
A Hugging Face dataset of Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings built from xSID's English alarm, reminder, and weather requests, shaped as When2Call multiple-choice decisions for post-training CLM heads.
Jev Models & Reimplementations#huggingface#open-models#benchmark
Community
A Hugging Face dataset of Qwen3.6-35B-A3B-FP8 embeddings for ToolRet training records, converted into Jev Decision Index ToolRet-style query-to-document decisions for post-training CLM heads. Ships float16 state, action, and hard-negative vectors plus per-row metadata.
Jev Models & Reimplementations#open-models#huggingface#jev-decision-index
Community
Embedding dataset from clm-project that renders NVIDIA's When2Call training split into Jev Decision Index-style tool-use decisions and stores Qwen3.6-35B-A3B-FP8 state, action, and hard-negative vectors for post-training CLM heads.
Jev Models & Reimplementations#open-models#research#agents
Community
Full-parameter fine-tune of Qwen3.5-4B for medical multiple-choice scoring. Given a case and candidate options, it returns the chosen option key with per-option probabilities as short JSON, without explanations or chain-of-thought.
Jev Models & Reimplementations#open-models#classification#python
Community
Hugging Face model card for JEV-Computer-Use, a System-One decision model for computer and browser use that reads out calibrated discrete action probabilities in a single forward pass instead of verbose text reasoning.
Jev Models & Reimplementations#open-models#browser#agents
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.