15 Downloads Updated 2 months ago
ollama run aevgeniou/qwen-cline
Updated 2 months ago
2 months ago
05df5cddb149 · 1.9GB
Qwen2.5-Coder-3B-Instruct (Q4_K_M) with lowered-temperature defaults for more deterministic code generation. Weights are unmodified.
| Tag | num_ctx | |
|---|---|---|
:16k |
16384 | same as :latest |
:8k |
8192 | lower memory footprint |
:latest |
16384 | default |
All tags share identical weights and differ only in num_ctx. Common to every tag: num_keep 24 · temperature 0.2 · top_p 0.9.
You do not strictly need the :8k tag — num_ctx is overridable at runtime with /set parameter num_ctx 8192, or via the options field in an API call. The tag exists as a convenience preset.
Note that Ollama’s model table reports 32K context for every tag; that is the architecture’s maximum, not the tag’s num_ctx default.
⚠️ Not an agentic model. Despite the “-cline” name, 3B cannot reliably hold tool-call protocols — use it for chat and code completion, not as an autonomous coding agent.
Built with Qwen. Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright © Alibaba Cloud. All Rights Reserved. NON-COMMERCIAL USE ONLY (research or evaluation). Commercial use requires a separate license from Alibaba Cloud.