Qwen2 / Qwen2.5

Get it

repo

Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF

file

qwen2.5-coder-0.5b-instruct-q4_k_m.gguf

quant

q4_k_m

size

0.49 GB

sha256

1d9614638d18024d…

Good for, and what it needs

Code completion and small edits; the smallest checkpoint here that writes usable Go.

~1 GB resident at --quant int4; runs on any laptop

Tool calling

minimal schema: ok; harness-scale (12 tools): skip — too small (measured 2026-09-07, nobara-pc)

Decisions · /v1/systemone

Label scoring, unmeasured.

What hasn't been shown

• Too small for an agent's full tool list: given twelve tools, it answered instead of calling one.
• Never measured on a machine like yours, so no speed is shown.

How sure we are

full-oracle 100.0%/1.00000 · what parity-gated means

Measured speed · decode, tokens per second

Ollama v0.32.5 at its defaults, same machine, same session. 128 tokens of context unless noted. Full method in benchmarks.

Get it

repo

Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF

file

qwen2.5-coder-1.5b-instruct-q4_k_m.gguf

quant

q4_k_m

size

1.1 GB

sha256

cc324af070c2ecbf…

Good for, and what it needs

Pinned as a release tier (pull/curated.json), not a registry checkpoint, so it carries no good-for, needs or tools fields.

Tool calling

not yet measured

Decisions · /v1/systemone

Label scoring, chat-v1, calibrated: top-1 0.2774, ECE 0.2102, over 872 out-of-distribution questions from a public decisions corpus. Measured 2026-09-28 on the RTX 2070 SUPER.

Label scoring reads the model's own probabilities for the option labels; it is not a trained decision head, and the record explains how far behind one it sits: the measurement.

What hasn't been shown

• Tool calling hasn't been measured on Qwen2.5-Coder 1.5B.
• Never measured on a machine like yours, so no speed is shown.

Measured speed · decode, tokens per second

Ollama v0.32.5 at its defaults, same machine, same session. 128 tokens of context unless noted. Full method in benchmarks.