binesh commited on
Commit
05e4e2c
·
verified ·
1 Parent(s): 06b4d96

Card: explicit scope note + link to what the experiments prove

Browse files
Files changed (1) hide show
  1. README.md +12 -9
README.md CHANGED
@@ -21,12 +21,20 @@ pretty_name: baby-MedGemma
21
  A small, **MedGemma-faithful probe** for causally isolating *where paraphrase
22
  sensitivity in medical vision-language models originates*: a frozen
23
  `google/medsiglip-448` encoder with a small **Gemma-3 decoder** (rotary position
24
- embeddings, RMSNorm, grouped-query attention: 6 query / 2 key-value heads),
25
- image tokens prepended inline. Loadable directly with the `transformers` API.
26
 
27
  - **Code, docs, and results:** https://github.com/thedatasense/babymedgemma
 
28
  - **Interactive write-up:** https://bineshkumar.me/phd-thesis/causality/
29
 
 
 
 
 
 
 
 
30
  ## Load and run
31
 
32
  ```python
@@ -67,13 +75,8 @@ narrow regime: Mann-Whitney U p = 1.4e-6, Cliff's delta = 1.00). A rank-1
67
  difference-direction patch restores flipped answers at the early layers with
68
  clean controls, so the flip is a low-rank, language-side, readout-stage
69
  direction. The origin is the training data; the fix is paraphrase augmentation.
70
-
71
- ## Intended use and limitations
72
-
73
- Research artifact for interpretability of paraphrase sensitivity in medical VLMs.
74
- It is a controlled probe, not the deployed MedGemma-4B, trained on ~1,800
75
- questions over a small word-level vocabulary; it grounds only weakly and
76
- manipulates adaptation-stage data. **Not a medical device; not for clinical use.**
77
 
78
  ## Training data and license
79
 
 
21
  A small, **MedGemma-faithful probe** for causally isolating *where paraphrase
22
  sensitivity in medical vision-language models originates*: a frozen
23
  `google/medsiglip-448` encoder with a small **Gemma-3 decoder** (rotary position
24
+ embeddings, RMSNorm, grouped-query attention: 6 query / 2 key-value heads), image
25
+ tokens prepended inline. Loadable directly with the `transformers` API.
26
 
27
  - **Code, docs, and results:** https://github.com/thedatasense/babymedgemma
28
+ - **What the experiments prove (a plain read):** https://github.com/thedatasense/babymedgemma/blob/main/docs/EXPERIMENTS.md
29
  - **Interactive write-up:** https://bineshkumar.me/phd-thesis/causality/
30
 
31
+ > **Scope.** This is a controlled research probe, **not a general medical VQA
32
+ > model**. It has a small domain-specific vocabulary (733 words from the
33
+ > chest-X-ray presence questions) and grounds weakly, so it is loadable and
34
+ > runnable for inspection and reproduction only. It supports a *sufficiency*
35
+ > claim about the cause of paraphrase sensitivity, not proof that the deployed
36
+ > MedGemma-4B has this exact origin. Not a medical device; not for clinical use.
37
+
38
  ## Load and run
39
 
40
  ```python
 
75
  difference-direction patch restores flipped answers at the early layers with
76
  clean controls, so the flip is a low-rank, language-side, readout-stage
77
  direction. The origin is the training data; the fix is paraphrase augmentation.
78
+ See [`docs/EXPERIMENTS.md`](https://github.com/thedatasense/babymedgemma/blob/main/docs/EXPERIMENTS.md)
79
+ for what each experiment does and does not prove.
 
 
 
 
 
80
 
81
  ## Training data and license
82