qwen2.5-coder:14b
Family: qwen2.5-coder · 14B parameters · 1 run(s)
Why this model scores about 0 on tool selection: what it wrote instead of tool calls, and whether the template, the engine or the model is the cause. Read the analysis. Applies to all runs on this page; scores unchanged.
Fits in 16 GB VRAM · ollama 0.40.6 · Q4_K_M
Measured VRAM at 32K context (KV cache included, ollama ps on the RTX 3090): 15 GB · previous rank in this tier: #8 (results v0.1.1, before the v0.1.2 re-tiering; score unchanged). Tier rule
24.2/100 · scoring v4
95% task-bootstrap CI (main suite, 3 trials): 21.2–27.0 · pooled over 8 runs (3 main-suite trials + 5 reruns at temperature 0.7): 23.4–28.2 · ≈ statistically tied with other models in pooled tie group 6 of this tier (the leaderboard’s ≈) · rank #15 in tier, rank range under weight sensitivity #15–#15 (5th–95th percentile; min #14, max #16) · v3: 24.2 (#13) · mention 0.0% (0/45) · how
Plain-language summary
Doesn't work as a tool-using agent in this suite.
Good for
- returning clean JSON
Weak at
- picking the right tool
- multi-step plans
- staying safe under prompt injection
Category breakdown
| JSON | Schema | Tool | Multi-step | Injection (useful&safe) | tok/s | p50 s/task | p95 s/task |
|---|---|---|---|---|---|---|---|
| 89 | 72 | 0 | 0 | 0 | 69.0 @b7e2c7 | 10.0 | 11.5 |
Prompt injection (15 tasks × 3 trials): useful and safe 0.0% (task completed and not compromised; counts in the score) · utility under attack 0.0% (completed the user's task) · harmful-action attack success 0.0% (a real harmful tool call or canary leak). Mentions (quoting the attacker without acting): 0.0% (0/45); they do not lower the score. v3 useful and safe: 0.0%. v2.1 used injection resistance 100.0 × tool-call rate 0. Definitions
- Run:
20261022-qwen2.5-coder-14b-c99868· 2026-10-22T23:45:15+00:00 - GPU: NVIDIA GeForce RTX 3090 (24 GB) · Linux 6.8.0-57-generic
- Host:
b7e2c79acebd· container (RunPod) · host AMD Ryzen Threadripper PRO 5995WX 64-Cores, 251.5 GB RAM · measured 2026-10-22T23:45:15+00:00 · hosts - Harness: hermes-agent Hermes Agent v0.22.2+2363.g9cb34e9 (2026.9.24) · stable · local 9cb34e9e285d · trials 3 · sampling: Ollama default (temperature and seed not set by harness) · context window: 32768 tokens as loaded by Ollama (the unused runner field records 8192; see Methodology)
- Score per trial (v4): 23.4 / 25.2 / 24 (SD 0.92, range 23.4–25.2)
- GGUF sha256:
ac9bc7a69dab38da1c790838955f1293420b55ab555ef6b4615efa1c1507b1ed - Sponsor disclosure: none
Chat template vs upstream
Matches: the chat template in the GGUF we tested is identical to the provider's upstream template after whitespace normalisation.
- GGUF template sha256 (normalised):
705d5782da398f5ce4bb10e19b2358aba0bd78710233d5f1811d7068cba1b58a - Upstream reference sha256 (normalised):
705d5782da398f5ce4bb10e19b2358aba0bd78710233d5f1811d7068cba1b58a· source
Labels are only “matches / differs / no reference”, never safe/unsafe. Ollama may use a built-in Go template or renderer instead of the GGUF Jinja. Why templates matter: Chat-template backdoors. Not part of the score.