File size: 3,632 Bytes
c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 c036088 93995c0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 | ---
tags:
- text-embeddings
- retrieval
- radiology
- chest
- qwen
base_model:
- Qwen/Qwen3-Embedding-4B
library_name: transformers
pipeline_tag: feature-extraction
---
# chest2vec_4B
Chest-radiology **text embedding** model: [`Qwen/Qwen3-Embedding-4B`](https://huggingface.co/Qwen/Qwen3-Embedding-4B)
contrastively LoRA-adapted for chest CT / CXR report retrieval. Embedding = left-padding-aware
last-token (EOS) pooling + L2-norm. **Embedding dim: 2560.**
## Self-contained `AutoModel`
The LoRA adapter is **merged into the weights** (`model.safetensors`) and the tokenizer is bundled,
so loading needs **no `chest2vec` package and no download of the base Qwen3-Embedding weights**:
```python
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("chest2vec/chest2vec_4B", trust_remote_code=True).eval()
tok = AutoTokenizer.from_pretrained("chest2vec/chest2vec_4B", trust_remote_code=True)
docs = ["Bibasilar atelectasis with small bilateral pleural effusions. Cardiomegaly."]
doc_emb = model.embed_texts(docs, tokenizer=tok) # [N, 2560], L2-normalized
# instruction-conditioned query
q_emb = model.embed_instruction_query(
"Retrieve the chest CT report that is similar to the given report.",
["pleural effusion and cardiomegaly"], tokenizer=tok)
vals, idx = model.cosine_topk(q_emb, doc_emb, k=5)
```
## Matryoshka embeddings
Matryoshka (MRL)-trained — truncate to **512** or **256** dims (keep first *N* dims, re-normalize):
```python
emb512 = model.embed_texts(docs, tokenizer=tok, dim=512)
emb256 = model.embed_texts(docs, tokenizer=tok, dim=256)
```
Recommended dims: **2560 (full) · 512 · 256** (`config.matryoshka_dims`). Use the same `dim` for query and corpus.
## Recommended instructions
Instruction-conditioned (`Instruct: {instruction}\nQuery: {report}`). Apply to the **query** side;
embed the corpus without an instruction. Trained on chest **CT and CXR** across these families:
**Retrieval** — `Retrieve the chest CT report that is similar to the given report.` ·
`Retrieve the CXR report that is similar to the given report.` ·
`Retrieve the CXR report that is similar to the given report with prior reference omitted.`
**Summarization** — `Summarize the following chest CT report` · `Summarize the following CXR report` · `Summarize the given report.`
**Entity extraction (leaf)** — `Given the following chest CT report, extract the presence/absence of entities` · `Given the following CXR report, extract the presence/absence of entities`
**Entity extraction (upper/coarse)** — `Given the following chest CT report, extract the presence/absence of upper-level entities` · `Given the following CXR report, extract the presence/absence of upper class entities`
**Anatomy-specific** — `From the following chest {CT report | X-ray report}, extract and return only the findings related to {REGION}, ignoring all information about other structures.`
- CT regions: lungs · airways and trachea · pleura · mediastinum and hilum · cardiovascular system · chest wall · bones and spine · upper abdomen · lower neck
- CXR regions: lungs and airways · pleura · hila and mediastinum · cardiovascular system · musculoskeletal structures and chest wall · tubes, catheters, and support devices · abdomen
## Details
- **Base:** Qwen/Qwen3-Embedding-4B (Apache-2.0) — architecture rebuilt from the bundled config; merged weights loaded from this repo. Default attention `sdpa` (use `flash_attention_2` on Ampere+ for speed).
- Merged weights reproduce the original adapter-based embeddings to **cosine ≥ 0.999**.
|