Datasets:

domofon
/

Gasai-Agent-CoderForge

license: other
language:
  - en
task_categories:
  - text-generation
tags:
  - agentic
  - software-engineering
  - tool-use
  - code
  - distillation
  - gasai
  - slm
size_categories:
  - 1K<n<10K
source_datasets:
  - togethercomputer/CoderForge-Preview
pretty_name: Gasai-Agent-CoderForge
configs:
  - config_name: default
    data_files: train.jsonl

Gasai-Agent-CoderForge

5,205 verified-resolved agentic software-engineering trajectories, re-authored into the Gasai harness format for pretraining small language models. Derived from togethercomputer/CoderForge-Preview (reward==1, Python). Each row is one complete tool-use trace serialized as a single Gasai control-token sequence in the gasai field.

Sister dataset to GasaiAI/Gasai-Agent-5k (from nvidia/Open-SWE-Traces) — byte-identical format, same pipeline.

What it is


Traces	5,205
Unique problems (instances)	2,704 (≤2 rollouts kept per instance)
Total tokens (StarCoder-2 tok)	~312M
Tokens / trace	median ~60K
File	`train.jsonl` (~1.0 GB)
Code language	Python
Natural language	English

Format (Gasai harness)

<|bos|><|system|>{system}<|tools|>[{read,bash,edit,write}]
<|user|>{structured task}
<|assistant|><think>{dense reasoning}</think><|tool_use|>{"name","input"}<|eos|>
<|tool_result|>{"tool_use_id":"tc_001","name","content"}<|eos|>
… (tool loop) …
<|assistant|><think>{reasoning}</think>{final answer}<|eos|>

One <think> per assistant turn; <|tool_use|>↔<|tool_result|> paired by positional tc_NNN. Schema: {"trajectory_id": "...", "gasai": "<|bos|>…<|eos|>"}.

Pipeline

Tool remap — CoderForge's OpenHands tools (execute_bash, str_replace_editor, finish, think) → the fixed Gasai 4-tool vocab read / bash / edit / write.
Constitution <think> — the teacher's verbose reasoning re-written per turn into a dense, source-bound <think> (every line a restated fact / grounded decision; no meta-narrative). Median ~42 tokens/think, ~0.01% task-restatement.
Structured prompts — first user turn restructured into a clean XML prompt (role/<context>, data in tags, request last, embedded data in {{ }}). 80% restructured; 20% keep the verbatim original (fidelity fallback, never lossy).
Deterministic polish — directory reads → bash find; dropped persistent-shell artifacts (C-c/empty bash, hung "command is still running" steps); coerced object-typed edit/write args; stripped emoji from finals (arrows kept); kept (incomplete) empty-final traces.
Dedup — ≤2 rollouts per base instance (the source has up to 8 reward==1 rollouts per problem).

⚠️ Known limitations (read before training)

This dataset was deep-audited (multi-agent semantic review). It is pretrain-grade as one ingredient, not a clean SFT set. Honest caveats:

Benchmark contamination — NOT decontaminated. Built from SWE-bench evaluation-harness images (sweb.eval.x86_64.*); upstream mixes SWE-smith / SWE-rebench / R2E-Gym (the latter two are live leaderboards). The 2,704 instances were not filtered against SWE-bench Verified/Lite or SWE-Gym test. Do not benchmark a model trained on this against those suites without first intersecting & removing overlapping instance_ids (it would be train-on-test).
<think> result-reading defect (~20–35% of traces, in-loss). Some <think> blocks state "the previous test failed" right after a passing run — the pytest N passed summary is buried in walls of = and the distiller mis-read it. Affects the result-reading skill; fixable by regenerating the affected think blocks.
Oracle-test reliance (inherent). Agents read the failing test, then implement exactly to it — valid TDD, reward is honest, but the in-loss think leans on a test oracle that won't exist on real tasks, so the apparent problem-solving capability is inflated.
Limited diversity. ~312M tokens come from only 2,704 unique problems (×≤2 rollouts); volume ≠ problem diversity. Mix with other sources for breadth.
Minor: ~7% of traces edit test files (a "fix the test" anti-pattern); first step is uniformly bash find /testbed -maxdepth 2; observations carry OpenHands wrappers ([exit code], cat -n numbering) and /testbed paths (conditioning only — tool observations are not in loss).

Source & license

Derived from togethercomputer/CoderForge-Preview, config filtered_reward1 (the curated reward==1 union of SWE-smith / SWE-rebench / R2E-Gym tasks). Underlying repositories carry their own permissive licenses (MIT / Apache-2.0 / BSD; see the per-row license in the source). Please cite CoderForge-Preview and the upstream task sets.