PEFT
Safetensors
Chinese
kev
queue-position
decision-model
lora
sft
permutation-consistency
chinese
Eval Results (legacy)
Instructions to use fzkun/kev-queue-v4-4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use fzkun/kev-queue-v4-4b with PEFT:
from peft import PeftModel from transformers import AutoModel base_model = AutoModel.from_pretrained("Qwen/Qwen3.5-4B-Base") model = PeftModel.from_pretrained(base_model, "fzkun/kev-queue-v4-4b") - Notebooks
- Google Colab
- Kaggle
Download reports/verification.json from fzkun/kev-queue-v4-4b: direct link, hf CLI and curl.
- Browser
- Download file 2.22 kB
-
https://huggingface.co/fzkun/kev-queue-v4-4b/resolve/main/reports/verification.json
- Command line
-
hf download hf://fzkun/kev-queue-v4-4b/reports/verification.json
-
curl -L -o verification.json https://huggingface.co/fzkun/kev-queue-v4-4b/resolve/main/reports/verification.json
2.22 kB
| { | |
| "health": { | |
| "status": "ok", | |
| "device": "cuda", | |
| "max_state_tokens": 8192, | |
| "checkpoint": "/queue-v4", | |
| "temperature": 0.25, | |
| "model": "kev-queue-v4" | |
| }, | |
| "model_card": { | |
| "name": "kev-queue-v4", | |
| "description": "Kev pointer head on Qwen/Qwen3.5-4B-Base, serving /queue-v4 at temperature 0.25", | |
| "release_date": "2026-10-05", | |
| "run": "/queue-v4", | |
| "base": "Qwen/Qwen3.5-4B-Base", | |
| "lora": 16, | |
| "device": "cuda", | |
| "backend": "torch", | |
| "dtype": "bfloat16", | |
| "temperature": 0.25, | |
| "max_state_tokens": 8192, | |
| "truncate_states": false, | |
| "cuda_graphs": null, | |
| "prefix_cache": { | |
| "size": 0, | |
| "min_state_tokens": 0, | |
| "max_tokens": 8192, | |
| "hits": 0, | |
| "misses": 0, | |
| "cached_states": 0, | |
| "oom_retries": 0 | |
| }, | |
| "batches": { | |
| "count": 1, | |
| "requests": 1, | |
| "queued": 0 | |
| } | |
| }, | |
| "gateway": "http://10.8.9.101:28080", | |
| "upstream": "http://192.168.0.81:8093", | |
| "api_key_reused": true, | |
| "n": 184, | |
| "seconds": 39.13232755661011, | |
| "offline_position_matches": 184, | |
| "max_probability_delta": 4.497899499256164e-05, | |
| "groups": { | |
| "blood_priority": { | |
| "correct": 25, | |
| "n": 25 | |
| }, | |
| "community_priority": { | |
| "correct": 25, | |
| "n": 25 | |
| }, | |
| "return_visit": { | |
| "correct": 29, | |
| "n": 30 | |
| }, | |
| "workflow_untrained": { | |
| "correct": 8, | |
| "n": 12 | |
| } | |
| }, | |
| "errors": [ | |
| { | |
| "id": "return_visit-random-19", | |
| "expected": 15, | |
| "predicted": 16 | |
| }, | |
| { | |
| "id": "return_visit-random-20/reverse", | |
| "expected": 24, | |
| "predicted": 25 | |
| }, | |
| { | |
| "id": "workflow-07", | |
| "expected": 3, | |
| "predicted": 4 | |
| }, | |
| { | |
| "id": "workflow-07/reverse", | |
| "expected": 3, | |
| "predicted": 4 | |
| }, | |
| { | |
| "id": "workflow-08", | |
| "expected": 7, | |
| "predicted": 10 | |
| }, | |
| { | |
| "id": "workflow-10", | |
| "expected": 2, | |
| "predicted": 10 | |
| }, | |
| { | |
| "id": "workflow-11", | |
| "expected": 5, | |
| "predicted": 10 | |
| } | |
| ], | |
| "upstream_gateway_sample_equal": true, | |
| "gateway_model": "jev-latest", | |
| "direct_queue_alias_equal": true, | |
| "unauthorized_status": 401, | |
| "over_limit_status": 422 | |
| } |