Spaces:
Running
Running
Document and map Gemini model config
Browse files
README.md
CHANGED
|
@@ -46,6 +46,50 @@ HuggingMess runs [Nous Research Hermes Agent](https://github.com/NousResearch/he
|
|
| 46 |
| `CLOUDFLARE_WORKERS_TOKEN` | Optional | Auto-creates a Worker proxy for Telegram Bot API traffic |
|
| 47 |
| `UPTIMEROBOT_API_KEY` | Optional | Auto-creates a monitor for `/health` |
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
## Telegram on HF Spaces
|
| 50 |
|
| 51 |
When `TELEGRAM_BOT_TOKEN` and `SPACE_HOST` are present, HuggingMess defaults Telegram to webhook mode:
|
|
|
|
| 46 |
| `CLOUDFLARE_WORKERS_TOKEN` | Optional | Auto-creates a Worker proxy for Telegram Bot API traffic |
|
| 47 |
| `UPTIMEROBOT_API_KEY` | Optional | Auto-creates a monitor for `/health` |
|
| 48 |
|
| 49 |
+
## LLM Providers
|
| 50 |
+
|
| 51 |
+
HuggingMess supports two configuration styles:
|
| 52 |
+
|
| 53 |
+
1. **Simple wrapper style:** set `LLM_MODEL` and `LLM_API_KEY`; HuggingMess maps them into Hermes config and provider-specific API key variables.
|
| 54 |
+
2. **Native Hermes style:** set Hermes variables directly, such as `HERMES_MODEL`, `HERMES_INFERENCE_PROVIDER`, and the provider API key.
|
| 55 |
+
|
| 56 |
+
### Gemini
|
| 57 |
+
|
| 58 |
+
For Google Gemini, add these Space secrets:
|
| 59 |
+
|
| 60 |
+
| Secret | Value |
|
| 61 |
+
| :--- | :--- |
|
| 62 |
+
| `LLM_MODEL` | `google/gemini-2.5-flash` |
|
| 63 |
+
| `LLM_API_KEY` | Your Google AI Studio API key |
|
| 64 |
+
|
| 65 |
+
HuggingMess will convert that into Hermes config:
|
| 66 |
+
|
| 67 |
+
```yaml
|
| 68 |
+
model:
|
| 69 |
+
default: gemini-2.5-flash
|
| 70 |
+
provider: gemini
|
| 71 |
+
```
|
| 72 |
+
|
| 73 |
+
And it exports both `GOOGLE_API_KEY` and `GEMINI_API_KEY` for Hermes. You can also use `gemini/gemini-2.5-flash`; the prefix is stripped the same way.
|
| 74 |
+
|
| 75 |
+
Native Hermes equivalent:
|
| 76 |
+
|
| 77 |
+
| Secret | Value |
|
| 78 |
+
| :--- | :--- |
|
| 79 |
+
| `HERMES_MODEL` | `gemini-2.5-flash` |
|
| 80 |
+
| `HERMES_INFERENCE_PROVIDER` | `gemini` |
|
| 81 |
+
| `GOOGLE_API_KEY` or `GEMINI_API_KEY` | Your Google AI Studio API key |
|
| 82 |
+
|
| 83 |
+
### Common Examples
|
| 84 |
+
|
| 85 |
+
| Provider | Simple `LLM_MODEL` | API key secret |
|
| 86 |
+
| :--- | :--- | :--- |
|
| 87 |
+
| Gemini | `google/gemini-2.5-flash` | `LLM_API_KEY` |
|
| 88 |
+
| OpenRouter | `openrouter/anthropic/claude-sonnet-4` | `LLM_API_KEY` |
|
| 89 |
+
| Anthropic | `anthropic/claude-opus-4.6` | `LLM_API_KEY` |
|
| 90 |
+
| OpenAI | `openai/gpt-4o` | `LLM_API_KEY` |
|
| 91 |
+
| Hugging Face Router | `huggingface/meta-llama/Llama-3.3-70B-Instruct` | `LLM_API_KEY` |
|
| 92 |
+
|
| 93 |
## Telegram on HF Spaces
|
| 94 |
|
| 95 |
When `TELEGRAM_BOT_TOKEN` and `SPACE_HOST` are present, HuggingMess defaults Telegram to webhook mode:
|
start.sh
CHANGED
|
@@ -124,6 +124,8 @@ case "$MODEL_PREFIX" in
|
|
| 124 |
;;
|
| 125 |
google|gemini)
|
| 126 |
[ -n "$LLM_API_KEY" ] && export GOOGLE_API_KEY="${GOOGLE_API_KEY:-$LLM_API_KEY}" GEMINI_API_KEY="${GEMINI_API_KEY:-$LLM_API_KEY}"
|
|
|
|
|
|
|
| 127 |
;;
|
| 128 |
deepseek)
|
| 129 |
[ -n "$LLM_API_KEY" ] && export DEEPSEEK_API_KEY="${DEEPSEEK_API_KEY:-$LLM_API_KEY}"
|
|
|
|
| 124 |
;;
|
| 125 |
google|gemini)
|
| 126 |
[ -n "$LLM_API_KEY" ] && export GOOGLE_API_KEY="${GOOGLE_API_KEY:-$LLM_API_KEY}" GEMINI_API_KEY="${GEMINI_API_KEY:-$LLM_API_KEY}"
|
| 127 |
+
[ "$PROVIDER_FOR_CONFIG" = "auto" ] && PROVIDER_FOR_CONFIG="gemini"
|
| 128 |
+
MODEL_FOR_CONFIG="${MODEL_INPUT#*/}"
|
| 129 |
;;
|
| 130 |
deepseek)
|
| 131 |
[ -n "$LLM_API_KEY" ] && export DEEPSEEK_API_KEY="${DEEPSEEK_API_KEY:-$LLM_API_KEY}"
|