257 Downloads Updated 5 days ago
A compact, agent-tuned Qwen 3.5 model built for local agent harnesses like Hermes Agent. Based on `qwen3.5:latest` with Q4_K_M quantization for a lightweight 6.6 GB footprint, while retaining strong tool-calling, vision, and thinking capabilities.
ollama run edtorre/qwen3.5-hermes
Applications
ollama launch claude --model edtorre/qwen3.5-hermes
ollama launch codex-app --model edtorre/qwen3.5-hermes
ollama launch openclaw --model edtorre/qwen3.5-hermes
ollama launch hermes --model edtorre/qwen3.5-hermes
ollama launch codex --model edtorre/qwen3.5-hermes
ollama launch opencode --model edtorre/qwen3.5-hermes
Models
View all →Readme
A compact, agent-tuned Qwen 3.5 model built for local agent harnesses like Hermes Agent. Based on qwen3.5:latest with Q4_K_M quantization for a lightweight 6.6 GB footprint, while
retaining strong tool-calling, vision, and thinking capabilities.
## Specs
| Parameter | Value |
|—————–|————————————|
| Base Model | qwen3.5:latest |
| Architecture | Qwen 3.5 |
| Parameters | 9.7B |
| Quantization | Q4_K_M |
| Size | 6.6 GB |
| Context Window | 65,536 (64K) |
| Max Output | 8,192 tokens |
| Temperature | 0.7 |
| Top-k / Top-p | 20 / 0.9 |
| Presence Penalty| 1.2 |
## Capabilities
- ✅ Tools / function calling
- ✅ Vision
- ✅ Audio
- ✅ Thinking
## Performance
Designed to run alongside these Ollama server settings:
OLLAMA_FLASH_ATTENTION=1
OLLAMA_KV_CACHE_TYPE=q8_0
At only 6.6 GB with flash attention + Q8_0 KV cache, total VRAM usage at 64K context is around ~12-13 GB — very comfortable on 16-20 GB GPUs, and even feasible on some 12 GB cards.
## Quick Start
ollama run edtorre/qwen3.5-hermes
With Hermes Agent:
hermes config set model.default edtorre/qwen3.5-hermes
hermes chat
License
Apache 2.0