Commands

Run (Exactly This)

uv pip install llama-cpp-python --no-cache

PowerShell

$env:CMAKE_ARGS="-DLLAMA_CUDA=off -DLLAMA_METAL=off"
uv pip install llama-cpp-python --no-cache

CMD

set CMAKE_ARGS=-DLLAMA_CUDA=off -DLLAMA_METAL=off
uv pip install llama-cpp-python --no-cache

Phase 1 Verification Results

PDF Processed: Academic PDF

Log Summary:

• Total chunks processed: [N]
• Total tokens: [N]
• Average tokens/second: [X]
• All three tasks logged: Yes/No

Sample Output:

• Summary: [paste full output from chunk 0]
• Concept: [paste full output from chunk 0]
• Chat Q&A: [paste full Q&A]

Phase 1 Status: ✅ COMPLETE

Criterion	Verification Method	Status
Chat works locally	chat_with_pdf() returns answers	✅
Summaries generated	summary_prompt_template produces output	✅
Concepts extracted	concept_prompt_template produces output	✅
No external API calls	Check log_df["hardware"] shows local CPU/GPU	✅
Cost/latency logged	experiments/phase1_logs.csv exists with data	✅