Datasets:
Add v5 and CrossNER results
Browse files
README.md
CHANGED
|
@@ -44,21 +44,30 @@ traces discuss the old output format.
|
|
| 44 |
| `openvocab_train_v4_labelonly.jsonl` | 3,131 | Identical to v4 with an **empty** `<think>` block: the control that isolates the effect of the reasoning. |
|
| 45 |
| `openvocab_train_v4_rejected.jsonl` | 3,131 | v4 recipe applied to CoTs **rejected** by the jury (same size as v4). |
|
| 46 |
| `openvocab_train_v4_all.jsonl` | 10,901 | v4 recipe on **all** usable CoTs, without the jury (3,131 accepted + 7,770 rejected). |
|
|
|
|
|
|
|
| 47 |
|
| 48 |
No file contains rows of the in-domain test set `universalmedner/pilener-eval-subset`.
|
| 49 |
|
| 50 |
-
## Main results (F1 type-agnostic, mean over
|
| 51 |
|
| 52 |
-
| Model | Biomedical | Pile-NER |
|
| 53 |
-
|---|---|---|
|
| 54 |
-
| Qwen3-4B base (thinking) | **0.602** | 0.573 |
|
| 55 |
-
| Qwen3-4B SFT v4 | 0.535 | 0.608 |
|
| 56 |
-
| Qwen3-4B SFT v4 label-only | 0.500 | **0.645** |
|
| 57 |
-
| Qwen3-
|
| 58 |
-
|
|
| 59 |
-
|
|
| 60 |
-
| Qwen3-1.7B
|
| 61 |
-
| Qwen3-1.7B SFT v4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
Training: LoRA r=16, alpha=32, lr 1e-4, 2 epochs (unless noted), loss on the
|
| 64 |
assistant turn only.
|
|
|
|
| 44 |
| `openvocab_train_v4_labelonly.jsonl` | 3,131 | Identical to v4 with an **empty** `<think>` block: the control that isolates the effect of the reasoning. |
|
| 45 |
| `openvocab_train_v4_rejected.jsonl` | 3,131 | v4 recipe applied to CoTs **rejected** by the jury (same size as v4). |
|
| 46 |
| `openvocab_train_v4_all.jsonl` | 10,901 | v4 recipe on **all** usable CoTs, without the jury (3,131 accepted + 7,770 rejected). |
|
| 47 |
+
| `openvocab_train_v5_teacher_qwen_3131.jsonl` | 3,131 | **v5**: v4 recipe, but the answer is the teacher's OWN entities (produced at the end of the same reasoning, as in ReasoningNER), so reasoning and answer never contradict each other; requested types = the teacher's labels in UPPER_SNAKE (e.g. `MEDICAL_CONDITION`); rows with a teacher entity not verbatim in the text are dropped. |
|
| 48 |
+
| `openvocab_train_v5_teacher_medgemma_3131.jsonl` | 3,131 | Same examples as v5 qwen, in MedGemma's reasoning format. |
|
| 49 |
|
| 50 |
No file contains rows of the in-domain test set `universalmedner/pilener-eval-subset`.
|
| 51 |
|
| 52 |
+
## Main results (F1 type-agnostic, mean over each test set)
|
| 53 |
|
| 54 |
+
| Model | Biomedical (9) | Pile-NER | CrossNER (5) |
|
| 55 |
+
|---|---|---|---|
|
| 56 |
+
| Qwen3-4B base (thinking) | **0.602** | 0.573 | **0.791** |
|
| 57 |
+
| Qwen3-4B SFT v4 | 0.535 | 0.608 | 0.751 |
|
| 58 |
+
| Qwen3-4B SFT v4 label-only | 0.500 | **0.645** | 0.716 |
|
| 59 |
+
| Qwen3-4B SFT v5 | 0.534 | 0.417 | 0.769 |
|
| 60 |
+
| MedGemma-1.5-4B base | 0.522 | 0.469 | 0.746 |
|
| 61 |
+
| MedGemma-1.5-4B SFT v5 | 0.527 | 0.436 | 0.783 |
|
| 62 |
+
| Qwen3-1.7B base (thinking) | 0.557 | 0.492 | – |
|
| 63 |
+
| Qwen3-1.7B SFT v4 | 0.487 | 0.567 | – |
|
| 64 |
+
| Qwen3-1.7B SFT v4, 1 epoch | 0.510 | 0.550 | – |
|
| 65 |
+
| Qwen3-1.7B SFT v4 rejected | 0.446 | 0.598 | – |
|
| 66 |
+
| Qwen3-1.7B SFT v4 all | 0.484 | 0.621 | – |
|
| 67 |
+
|
| 68 |
+
MedGemma v5 improves its base on 11 of 15 test sets, significantly on CrossNER (+0.037, 95% CI
|
| 69 |
+
[+0.027, +0.047]); on the biomedical mean it is on par (AnatEM drops, the teacher treats generic
|
| 70 |
+
anatomical nouns as non-entities). Adapters: `universalmedner/*-lora`.
|
| 71 |
|
| 72 |
Training: LoRA r=16, alpha=32, lr 1e-4, 2 epochs (unless noted), loss on the
|
| 73 |
assistant turn only.
|