Skip to content

Add Groq provider documentation, example script, and BrowseComp-Plus launcher - #4

Open
senthilkumaranT wants to merge 1 commit into
DCI-Agent:mainfrom
senthilkumaranT:groq-provider
Open

Add Groq provider documentation, example script, and BrowseComp-Plus launcher#4
senthilkumaranT wants to merge 1 commit into
DCI-Agent:mainfrom
senthilkumaranT:groq-provider

Conversation

@senthilkumaranT

Copy link
Copy Markdown

Summary

Adds end-to-end documentation and runnable scripts so DCI-Agent-Lite can use Groq as a Pi provider (OpenAI-compatible API), without changing existing defaults or benchmark judging.

Changes

  • assets/docs/groq.mdGROQ_API_KEY, sample Pi models.json groq provider (https://api.groq.com/openai/v1), run commands, pointer to Groq OpenAI compatibility.
  • scripts/examples/dci_basic_groq_example.sh — basic wiki-corpus example with --provider groq (model id must match Pi config).
  • scripts/bcplus_eval/run_bcplus_eval_groq.sh — BrowseComp-Plus batch eval launcher mirroring the OpenAI script; outputs under outputs/bcplus_eval/groq_....
  • .env.template — optional GROQ_API_KEY line.
  • src/dci/benchmark/pi_rpc_runner.py--provider help text mentions custom slugs (e.g. groq) and links to the doc.
  • Docs / README — cross-links in setup.md, benchmark.md, running.md, cli-reference.md, and README.md.

Out of scope / unchanged

  • Default agent provider for existing eval entrypoints (e.g. run_bcplus_eval.py defaults) unchanged unless callers use the new Groq launcher.
  • BrowseComp-Plus grading still uses the OpenAI judge (OPENAI_API_KEY).

Test plan

  • Configure Pi models.json per assets/docs/groq.md, set GROQ_API_KEY, run bash scripts/examples/dci_basic_groq_example.sh.
  • (Optional) Run bash scripts/bcplus_eval/run_bcplus_eval_groq.sh with a small --limit if you use it for a smoke test; ensure OPENAI_API_KEY is set for judging.

Document Groq as an OpenAI-compatible Pi provider (models.json, GROQ_API_KEY) in assets/docs/groq.md.

Add scripts/examples/dci_basic_groq_example.sh and scripts/bcplus_eval/run_bcplus_eval_groq.sh.

Update .env.template, dci-agent-lite --provider help, and cross-links in README and docs (setup, benchmark, running, cli-reference).

Defaults and BrowseComp-Plus judging (OpenAI) are unchanged.
charmandercha pushed a commit to charmandercha/DCI-Agent-Lite that referenced this pull request May 19, 2026
Correções aplicadas (ordenadas por esforço):

DCI-Agent#4  — Remove átomo duplicado cc_art205_atom1.md (>20 palavras, redundante)
DCI-Agent#6  — Gera _manifest.json para todas as 34 seções de atomos
DCI-Agent#3  — Adiciona validador 'Art. N' no gerar_atomos.py e aviso no revisar.sh
DCI-Agent#5  — Gera átomo do Art. 48-A (assembleias eletrônicas)
DCI-Agent#1  — Corrige slugs hierárquicos: 'disposicoes_gerais' dividido em 3 seções
       → pessoas_juridicas_disposicoes_gerais (arts 40-52)
       → negocio_juridico_disposicoes_gerais (arts 104-114)
       → prescricao_disposicoes_gerais (arts 189-196)
DCI-Agent#2  — Regenera 44 átomos sem referência de artigo em 4 seções
       → causas_que_interrompem_prescricao: +Art. 202 (8 átomos)
       → negocio_juridico_disposicoes_gerais: +Art. 104-114 (16 átomos)
       → invalidade_negocio_juridico: +Art. 166-184 (18 átomos)
       → lesao: +Art. 157 (2 átomos)

Resultados:
- 34:34 mapeamento 1:1 integra↔atomos (antes: 34:32)
- 0 átomos sem referência de artigo (antes: 44/249 = 17.7%)
- 0 átomos >20 palavras (antes: 1)
- sff precision@3: 60% (calibrado com dados reais)
- test_integracao_pipeline.sh: 8/8 PASS
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.

1 participant