Spaces:
Sleeping
Sleeping
Commit History
-
Tighten markdown->LaTeX converter; note SSH agent step in CLAUDE.md
- generate_latex.py: bold regex now tolerates embedded italic; bold rendering applies italic substitution on the captured content; drop horizontal rule lines; strip blockquote markers; commentary is now passed through md_to_latex so stray markdown from the LLM renders properly. Tightened the commentary prompt with an explicit no- markdown instruction. - CLAUDE.md: documented the ssh-agent load step needed before pushing to HF space2/space3. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>9b94583 -
Add multi-model comparison to run_examples; bump max_tokens to 700
- run_examples() now produces 13 results: 6 single-model examples (Gemini-2.5 Pro default) plus 7 multi-model runs for Witchweed Africa (one per supported LLM), enabling two-axis appendix coverage in the paper - generate_latex.py writes fragments into the examples/ subdirectory of the writing repo; handles single-# headers so Claude Opus-style titles render as \textbf instead of escaped \# - max_tokens bumped from 400 to 700 so longer multi-model responses (Claude Opus, FAW USA) don't truncate mid-sentence; prompt keeps the ~300-word soft cap in place Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>1c52b33 -
Auto-generate appendix examples; add Asian copperleaf to knowledge base
Workflow: run_examples() in app.py produces example_results_*.json, then generate_latex.py emits per-example LaTeX fragments into the writing repo that main.tex includes via \input{}. This keeps the paper in sync with the actual bot output. Changes: - Rewrote examples-work-out/generate_latex.py from scratch: reads latest JSON, emits one fragment per example, uses the paper's \configbox/ \querybox macros, converts markdown (bold, italic, nested bullets, headers) to LaTeX, and calls GPT-4.1 for 2-3 sentence grounded commentary using a prompt that spells out the paper's storyline - Updated app.py::run_examples(): six examples with stable ids, default model switched to Gemini-2.5 Pro, added species_display and id fields to result dicts - Added explicit word-count guidance to the RAG prompt template so responses stay within token limits and don't truncate mid-sentence - Swapped Sahara mustard for Asian copperleaf (Acalypha australis) in species_table.tex: Midwest USA tier -…609e744
-
Upgrade all LLM models to latest versions, add test harness
Models updated: GPT-3.5 -> GPT-4.1 Mini, Llama-3 -> Llama-4, Gemini-1.5 Pro -> Gemini-2.5 Pro, Claude 3 -> Claude 4.6. Default changed to Gemini-2.5 Pro. Fallback no longer exits. Added Gemini max_tokens buffer for total-token counting. Added test_app_models.py (7/7 models pass). Includes SLF Example 5 re-runs with GPT-4.1 and Gemini-2.5 Pro. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>5bbe904 -
Add SLF reporting sources to knowledge base, increase retriever k to 8
Added Iowa Dept of Agriculture and ISU Extension PDFs for spotted lanternfly reporting guidance. Rebuilt vector DB with clean sources. Increased retriever k from 3 to 8 to match paper's evaluation range and capture both identification and reporting chunks. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>e96f60e
-
Document vignette workflow in README, add latest example results
- Add "Example Vignettes → Paper Workflow" section to README - Clarify generate_latex.py is standalone (not used by main.tex) - Add example_results_20260314_023319.json with fresh vignette outputs - Add CLAUDE.md with project instructions Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>6a39926
-
Add example results output 20260303
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>5dc51d2 -
Add multi-page PDF support, expand USA tier to 219 species, add all 4 vignettes
- Fix app_database_prep to ingest all pages of multi-page PDFs (not just first page) - Expand USA IPM tier from 110 to 219 species - Add all 4 example vignettes: SLF, Striga, FAW, Sahara Mustard - Improve RAG prompt to be more concise and avoid document number references - Update README with chunk counts, new species pipeline docs, and updated eval metrics - Rebuild vector database and regenerate evaluation results Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>f8fa434 -
Update species verification note for bollworm replacement
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>1bc2d18
-
Add Spodoptera frugiperda examples for paper supplementary
- Switch example runner to fall armyworm (India + USA regions) - Add generated example results JSON - Update README with new example entry Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>cdfe190 -
Update docs, gitignore, examples, and vector DB
- Add LaTeX artifacts and editor temp files to .gitignore - Improve README with environment setup, examples/LaTeX commands, clearer DB build flow - Simplify example runner to Africa + USA regions only - Update species Excel and rebuild vector database Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>0198d9a
-
Simplify RAG prompt, add common name display, add example runner
- Fix dotenv loading with override=True in app.py and app_database_prep.py - Simplify the LLM prompt to reduce hallucination (remove complex conditionals and hardcoded references) - Add explicit region descriptions in prompt (Midwest USA, USA, Africa, India) - Display common name in species availability info - Add --examples flag to app.py for running predefined queries programmatically - Add generate_latex.py for converting example results to LaTeX - Rebuild vector database with all 4 regions (859 documents) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>9e21e50
-
start feb l2
5d805ea