init
Browse files
README.md
CHANGED
|
@@ -1,12 +1,36 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
-
colorTo:
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
| 8 |
-
|
| 9 |
-
short_description: codegen
|
| 10 |
---
|
| 11 |
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: Gemma 4 26B Coding API
|
| 3 |
+
emoji: 🚀
|
| 4 |
+
colorFrom: indigo
|
| 5 |
+
colorTo: green
|
| 6 |
sdk: docker
|
| 7 |
pinned: false
|
| 8 |
+
app_port: 7860
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
+
# Gemma 4 26B A4B — Coding API
|
| 12 |
+
|
| 13 |
+
Gemma 4 26B A4B (MoE) served as an Anthropic + OpenAI compatible API.
|
| 14 |
+
Model: `unsloth/gemma-4-26B-A4B-it-GGUF` · `UD-IQ3_XXS` (11.2 GB)
|
| 15 |
+
Params: temp=0.3, top_p=0.9, min_p=0.1, top_k=20 (coding-tuned)
|
| 16 |
+
|
| 17 |
+
## Claude Code setup
|
| 18 |
+
|
| 19 |
+
```bash
|
| 20 |
+
export ANTHROPIC_BASE_URL=https://YOUR_USERNAME-gemma-4-26b-coding-api.hf.space
|
| 21 |
+
export ANTHROPIC_API_KEY=gemma4-local
|
| 22 |
+
claude --model gemma-4-26b
|
| 23 |
+
```
|
| 24 |
+
|
| 25 |
+
## Environment variables (HF Space → Settings → Variables)
|
| 26 |
+
|
| 27 |
+
| Variable | Default | Description |
|
| 28 |
+
|----------|---------|-------------|
|
| 29 |
+
| `SPACE_URL` | `` | Your space URL — enables self-ping |
|
| 30 |
+
| `MODEL_REPO` | `unsloth/gemma-4-26B-A4B-it-GGUF` | HF repo |
|
| 31 |
+
| `MODEL_FILE` | `gemma-4-26B-A4B-it-UD-IQ3_XXS.gguf` | GGUF filename |
|
| 32 |
+
| `N_CTX` | `4096` | Context window |
|
| 33 |
+
| `N_THREADS` | `2` | CPU threads |
|
| 34 |
+
| `DEFAULT_TEMP` | `0.3` | Temperature |
|
| 35 |
+
| `DEFAULT_MIN_P` | `0.1` | Min-p (key for coding accuracy) |
|
| 36 |
+
| `DEFAULT_TOP_K` | `20` | Top-k |
|