Downloads · 30 days
0
hxgdzyuyi/qwen3-8b-steam-entity-linking-poc-b
qwen3-8b-steam-entity-linking-poc-b is a machine learning model from hxgdzyuyi. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as apache-2.0.
This repository contains only a low-rank residual projection and 1000 cosine class prototypes. The Qwen backbone is frozen and is loaded separately at its pinned revision. There is no RAG, vector database, candidate r…
Downloads · 30 days
0
Access
Public
Updated Aug 25, 2026
Repo size
217 MB
Likes
0
Public
Click a slice to open those files.
.safetensors247 MB · 94%
From the Hugging Face model README
This repository contains only a low-rank residual projection and 1000 cosine class prototypes. The Qwen backbone is frozen and is loaded separately at its pinned revision. There is no RAG, vector database, candidate retrieval, reranking, LoRA, or generated AppID token. The local feature cache and optimizer state are not part of this repository.
Qwen/Qwen3-8B-Base49e3418fbbbca6ecbdf9608b4d22e5a407081db4c4d33e6e7c435947e5856bd723814795b067404chttps://github.com/hxgdzyuyi/qwen-steam-entity-linking.git| Variant | Epoch | Canonical Top-1 | Alias Top-1 |
|---|---|---|---|
| published PoC A | 20 | 97.90% | 8.15% |
| zero-training prototypes | 0 | 84.80% | 14.13% |
| trained head | 1 | 17.10% | 2.72% |
| trained head | 3 | 91.60% | 19.57% |
| trained head | 5 | 99.10% | 30.43% |
| trained head | 10 | 99.90% | 42.93% |
| trained head | 20 | 100.00% | 46.20% |
Canonical acceptance is 95%. Alias is report-only.
Compared with the pinned published PoC A reference, alias_improved_over_poc_a
is true.
The loader reads classifier_config.json, downloads the exact Qwen revision,
freezes it, and attaches this repository's classifier head.
import sys
from huggingface_hub import snapshot_download
repo_dir = snapshot_download("hxgdzyuyi/qwen3-8b-steam-entity-linking-poc-b")
sys.path.insert(0, repo_dir)
from steam_entity_classifier import SteamEntityLinker
model = SteamEntityLinker.from_pretrained(repo_dir)
result = model.predict(["CS2", "反恐精英"], top_k=5)
print(result)
The Top-1 result contains appid, canonical_name, class_index, and
confidence; top_k is diagnostic.
This is a 1000-class closed-set classifier. It always emits one known AppID and
does not support UNKNOWN. Results for aliases, descriptions, and games newer
than the base model's knowledge cutoff may be weak. The 184 alias cases were
never used for training, checkpoint gating, or prototype initialization.