Skip to content

docs: replace SWE-Bench example with GSM8K math tutoring (en + zh) - #12

Merged
Protocol-zero-0 merged 2 commits into
mainfrom
docs/gsm8k-example
May 13, 2026
Merged

docs: replace SWE-Bench example with GSM8K math tutoring (en + zh)#12
Protocol-zero-0 merged 2 commits into
mainfrom
docs/gsm8k-example

Conversation

@Protocol-zero-0

Copy link
Copy Markdown
Owner

Summary

  • Replaces the "See it in action" example in both README.md and README.zh.md
  • Before: SWE-Bench Qwen3-Coder-7B, 32.4%→76.4%, code-repair benchmark
  • After: GSM8K Qwen3-7B-Instruct, 51.8%→96.2%, grade-school math word problems
  • Config reference mission, command, and allowed_paths updated to match in both READMEs

Why GSM8K

GSM8K is a better anchor example for this README:

  • Universal legibility — every parent, investor, and developer understands "a kid can't solve math word problems"
  • Before/after is concrete — 51.8% (worse than a 5th grader) → 96.2% (on par with GPT-5.5)
  • Market is obvious — tutoring / AI question-bank is a multi-hundred-billion-dollar market
  • Local inference angle — 7B model means any device, zero API cost, full privacy
  • 6 generations clearly motivated (CoT → self-consistency → calculator → verifier → decompose → best-of-16+verifier)

Test plan

  • English README renders correctly (badge links, code blocks, progress chart)
  • Chinese README is a faithful translation of the English version
  • No test changes — 39/39 tests unaffected (README is docs-only)

🤖 Generated with Claude Code

Swap "See it in action" section in both READMEs: SWE-Bench Qwen3-Coder-7B
→ GSM8K Qwen3-7B-Instruct (51.8%→96.2%, $34.10, 25 gens). Math tutoring
is a more universally legible value story — every parent understands the
market, the baseline, and the outcome.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 10, 2026 21:58

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Updates the project’s “See it in action” documentation examples to use a GSM8K math-tutoring benchmark narrative (EN + ZH) instead of the prior SWE-Bench example, and aligns the referenced config fields (mission, command, allowed_paths) across both READMEs.

Changes:

  • Replaces the README demo story/metrics from SWE-Bench (Qwen3-Coder-7B) to GSM8K (Qwen3-7B-Instruct), including updated generation-by-generation walkthroughs.
  • Updates the sample config snippets in both README.md and README.zh.md to reference the new mission/benchmark and mutation scope paths.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 6 comments.

File Description
README.md Replaces SWE-Bench example narrative + updates config snippet fields to GSM8K math-solver framing.
README.zh.md Mirrors the English README updates with a Chinese version of the GSM8K narrative and matching config snippet fields.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread README.md
Comment on lines +66 to +69
command: "python3 scripts/run_gsm8k.py --model qwen3-7b-instruct --sample 100 --json"

mutation_scope:
allowed_paths: ["src/agent_harness/"]
allowed_paths: ["src/math_solver_harness/"]
Comment thread README.md
Comment on lines 257 to 260
evidence_sources:
- type: shell # stdout goes into observation.json
command: "python3 scripts/run_benchmark.py --sample 50 --json"
command: "python3 scripts/run_gsm8k.py --model qwen3-7b-instruct --sample 100 --json"
- type: file # file contents go into observation.json
Comment thread README.md
mutation_scope:
allowed_paths:
- "src/agent_harness/" # changes outside this list are auto-rejected
- "src/math_solver_harness/" # changes outside this list are auto-rejected
Comment thread README.zh.md
Comment on lines +66 to +69
command: "python3 scripts/run_gsm8k.py --model qwen3-7b-instruct --sample 100 --json"

mutation_scope:
allowed_paths: ["src/agent_harness/"]
allowed_paths: ["src/math_solver_harness/"]
Comment thread README.zh.md
Comment on lines 257 to 260
evidence_sources:
- type: shell # stdout 写入 observation.json
command: "python3 scripts/run_benchmark.py --sample 50 --json"
command: "python3 scripts/run_gsm8k.py --model qwen3-7b-instruct --sample 100 --json"
- type: file # 文件内容写入 observation.json
Comment thread README.zh.md
mutation_scope:
allowed_paths:
- "src/agent_harness/" # 不在列表里的改动自动拒绝
- "src/math_solver_harness/" # 不在列表里的改动自动拒绝
Add a note above each config snippet explaining that scripts/run_gsm8k.py
and src/math_solver_harness/ are paths in the user's target project, not
this repo. Point to examples/evolution.yml for a runnable demo.

Addresses Copilot review comments on PR #12.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@Protocol-zero-0
Protocol-zero-0 merged commit 3cee524 into main May 13, 2026
2 checks passed
@Protocol-zero-0
Protocol-zero-0 deleted the docs/gsm8k-example branch May 13, 2026 14:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants