Views
No views yet
Disclaimer — research use only. The outputs are predicted class IDs from a frozen 1-of-K head built into a chat-style prompt; they are not clinical advice.
| Stage | Repo |
|---|---|
| Base | google/gemma-4-E4B-it |
| LoRA adapter (140 MB) | xxue752/mental_health_best_model |
| Merged bf16 (~16 GB) | xxue752/mental_health_best_model_merged |
| LiteRT INT4 (this repo) | xxue752/mental_health_best_model_litert |
| File | Size | Purpose |
|---|---|---|
model.litertlm | 4.12 GB | Single-file LiteRT-LM bundle (preferred) — load with litert-lm runtime |
model_quantized.tflite | 2.2 GB | Prefill + decode TFLite (dynamic_wi4_afp32) |
embedder_quantized.tflite | 327 MB | External token embedder (Gemma 4 requires externalize_embedder=True) |
per_layer_embedder_quantized.tflite | 1.4 GB | Per-Layer Embeddings (Gemma 4 PLE) |
tokenizer.json, tokenizer_config.json | 32 MB | Gemma 4 tokenizer |
chat_template.jinja | 17 KB | Chat template (LiteRT-LM-compatible variant — see "Compatibility fix" below) |
llm_metadata.pb | 17 KB | LiteRT-LM runtime metadata (carries the patched chat template) |