Parameter Floors for Developer-Agent Subroutines

ishaanranjan 's Collections

updated Jun 13

Full-parameter SFT checkpoints for eight deterministic, schema-verified coding-agent subroutines across 135M to 1.5B.

Upvote

ishaanranjan/slm-agent-action-router-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 19

Note Agent Action Router at SmolLM2 135M: 62.8% held-out success
ishaanranjan/slm-agent-action-router-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 26

Note Agent Action Router at SmolLM2 360M: 74.8% held-out success
ishaanranjan/slm-agent-action-router-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 18

Note Agent Action Router at Qwen2.5 0.5B: 96.0% held-out success
ishaanranjan/slm-agent-action-router-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 18

Note Agent Action Router at Qwen2.5 1.5B: 97.2% held-out success
ishaanranjan/slm-agent-evidence-judge-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 14

Note Code Evidence Judge at SmolLM2 135M: 53.6% held-out success
ishaanranjan/slm-agent-evidence-judge-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 11

Note Code Evidence Judge at SmolLM2 360M: 50.0% held-out success
ishaanranjan/slm-agent-evidence-judge-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 11

Note Code Evidence Judge at Qwen2.5 0.5B: 85.2% held-out success
ishaanranjan/slm-agent-evidence-judge-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 10

Note Code Evidence Judge at Qwen2.5 1.5B: 91.2% held-out success
ishaanranjan/slm-agent-json-repair-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 53

Note JSON Tool-Call Repair at SmolLM2 135M: 91.6% held-out success
ishaanranjan/slm-agent-json-repair-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 9

Note JSON Tool-Call Repair at SmolLM2 360M: 100.0% held-out success
ishaanranjan/slm-agent-json-repair-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 12

Note JSON Tool-Call Repair at Qwen2.5 0.5B: 100.0% held-out success
ishaanranjan/slm-agent-json-repair-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 13

Note JSON Tool-Call Repair at Qwen2.5 1.5B: 100.0% held-out success
ishaanranjan/slm-agent-path-normalizer-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 11

Note Repository Path Normalizer at SmolLM2 135M: 48.4% held-out success
ishaanranjan/slm-agent-path-normalizer-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 8

Note Repository Path Normalizer at SmolLM2 360M: 70.0% held-out success
ishaanranjan/slm-agent-path-normalizer-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 13

Note Repository Path Normalizer at Qwen2.5 0.5B: 94.4% held-out success
ishaanranjan/slm-agent-path-normalizer-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 10

Note Repository Path Normalizer at Qwen2.5 1.5B: 96.8% held-out success
ishaanranjan/slm-agent-read-span-selector-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 11

Note Code Read-Span Selector at SmolLM2 135M: 0.0% held-out success
ishaanranjan/slm-agent-read-span-selector-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 12

Note Code Read-Span Selector at SmolLM2 360M: 2.8% held-out success
ishaanranjan/slm-agent-read-span-selector-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 8

Note Code Read-Span Selector at Qwen2.5 0.5B: 21.6% held-out success
ishaanranjan/slm-agent-read-span-selector-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 9

Note Code Read-Span Selector at Qwen2.5 1.5B: 83.2% held-out success
ishaanranjan/slm-agent-search-hit-ranker-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 9

Note Code Search-Hit Ranker at SmolLM2 135M: 16.0% held-out success
ishaanranjan/slm-agent-search-hit-ranker-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 8

Note Code Search-Hit Ranker at SmolLM2 360M: 19.2% held-out success
ishaanranjan/slm-agent-search-hit-ranker-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 8

Note Code Search-Hit Ranker at Qwen2.5 0.5B: 94.0% held-out success
ishaanranjan/slm-agent-search-hit-ranker-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 13

Note Code Search-Hit Ranker at Qwen2.5 1.5B: 98.0% held-out success
ishaanranjan/slm-agent-search-query-gen-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 13

Note Code Search-Query Generator at SmolLM2 135M: 64.8% held-out success
ishaanranjan/slm-agent-search-query-gen-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 8

Note Code Search-Query Generator at SmolLM2 360M: 62.8% held-out success
ishaanranjan/slm-agent-search-query-gen-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 15

Note Code Search-Query Generator at Qwen2.5 0.5B: 88.4% held-out success
ishaanranjan/slm-agent-search-query-gen-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 8

Note Code Search-Query Generator at Qwen2.5 1.5B: 77.2% held-out success
ishaanranjan/slm-agent-trace-localizer-smollm2-135m

Text Generation • 0.1B • Updated Jun 13 • 10

Note Python Traceback Localizer at SmolLM2 135M: 80.8% held-out success
ishaanranjan/slm-agent-trace-localizer-smollm2-360m

Text Generation • 0.4B • Updated Jun 13 • 8

Note Python Traceback Localizer at SmolLM2 360M: 92.4% held-out success
ishaanranjan/slm-agent-trace-localizer-qwen2-5-0-5b

Text Generation • 0.5B • Updated Jun 13 • 9

Note Python Traceback Localizer at Qwen2.5 0.5B: 99.2% held-out success
ishaanranjan/slm-agent-trace-localizer-qwen2-5-1-5b

Text Generation • 2B • Updated Jun 13 • 9

Note Python Traceback Localizer at Qwen2.5 1.5B: 100.0% held-out success

Upvote