You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

PoC-SNOMEDCT-extract-gemma-4-e2b-it-medium

Proof-of-concept -- trained on synthetic data. Every run behind this repo so far fine-tunes on a synthetically generated batch, not real MIMIC-IV-Note. Treat these weights as a proof of concept, not a clinically validated model.

LoRA adapter for SNOMED-CT span extraction, fine-tuned from google/gemma-4-E2B-it.

Required system prompt

PROMPT_VARIANT = "medium". Inference must use this exact prompt — a different one silently degrades output:

(Omitted here.)

Run config


timestamp	2026-08-06 05:47:12
run_mode	train
model_id	google/gemma-4-E2B-it
gpu_target	L4
prompt_variant	medium
max_length	4096
fast_iter	False
n_train_chunks	144
n_train_notes	107
n_val_chunks	38
n_val_notes	27
epochs	4
train_batch	1
grad_accum	16
grad_ckpt	True
train_seconds	5818.1

Fine-tuned

Invalid output: 3/38 (7.9%)

Category + text (multiset)

category	P	R	F1
micro	0.289	0.170	0.214
macro	0.240	0.137	0.163
body_structure	0.000	0.000	0.000
finding	0.118	0.112	0.115
procedure	0.395	0.194	0.260

Positional (char-level)

category	P	R	F1	IOU
body_structure	0.000	0.000	0.000	0.000
finding	0.150	0.164	0.157	0.115
procedure	0.287	0.178	0.220	0.102

Downloads last month: 1

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for kiwiki196/PoC-SNOMEDCT-extract-gemma-4-e2b-it-medium

Base model

google/gemma-4-E2B

Finetuned

google/gemma-4-E2B-it

Adapter

(174)

this model