tests/integration/config_orc_fc_deep_research.py
Rama main · réplica de lectura del repositorio original.
import dotenv
dotenv.load_dotenv()
# hyperbolic/meta-llama/Llama-3.3-70B-Instruct
# hyperbolic/deepseek-ai/DeepSeek-V3
# hyperbolic/Qwen/Qwen3-235B-A22B
# hyperbolic/Qwen/Qwen2.5-VL-7B-Instruct
# hyperbolic/openai/gpt-oss-120b
# together-ai/deepcogito/cogito-v2-1-671b
# together-ai/ServiceNow-AI/Apriel-1.6-15b-Thinker
# together-ai/deepcogito/cogito-v2-preview-llama-109B-MoE <-- Model not available
# together-ai/deepcogito/cogito-v2-preview-llama-405B <-- model not available
# together-ai/deepcogito/cogito-v2-preview-llama-70B < -- model not available
# together-ai/deepseek-ai/DeepSeek-R1-0528-tput < -- model not available
# together-ai/deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free < -- model not available
# together-ai/google/gemma-2b-it-Ishan <---Input validation error: `inputs` tokens + `max_new_tokens` must be <= 8193
# together-ai/meta-llama/Llama-3-70b-hf < -- Not entities appropved list
# together-ai/meta-llama/Llama-3.1-405B-Instruct < -- Not in approved list
# together-ai/meta-llama/Llama-3.2-1B-Instruct < -- Not in approved list
DEEP_RESEARCH_TEST_SUITE = {
"test_1_deep_comparative": {
"name": "Deep Comparative (Multi-Metric)",
"query": (
"Compare NVIDIA and AMD's 2024 performance across:\n"
"1. Annual revenue\n"
"2. Data center segment revenue\n"
"3. Gaming segment revenue\n"
"4. Year-over-year growth rates\n\n"
"Cite official investor relations sources."
),
"expected_sources": 4, # 2 per company minimum
"complexity": "TIER_2",
"validates": ["multi_entity_research", "segment_breakdown", "source_citation"],
},
"test_2_trend_analysis": {
"name": "Trend Analysis (Temporal Depth)",
"query": (
"How has NVIDIA's data center revenue grown from 2021 to 2024? "
"Include quarterly breakdowns for 2024 and identify key inflection points."
),
"expected_sources": 5, # Multiple years + quarterly data
"complexity": "TIER_3",
"validates": [
"temporal_research",
"quarterly_granularity",
"trend_identification",
],
},
"test_3_multi_source_verification": {
"name": "Multi-Source Verification",
"query": (
"What was SpaceX's valuation in its most recent funding round? "
"Verify the answer from at least 3 different sources and note any discrepancies."
),
"expected_sources": 3, # Explicitly required
"complexity": "TIER_2",
"validates": ["source_diversity", "conflict_detection", "verification_depth"],
},
"test_4_complex_synthesis": {
"name": "Complex Synthesis (Multi-Vendor)",
"query": (
"What are the key differences between AWS, Azure, and Google Cloud's "
"AI infrastructure offerings as of Q4 2024? Focus on:\n"
"- GPU availability (H100, A100)\n"
"- Pricing per compute hour\n"
"- Integration with major ML frameworks"
),
"expected_sources": 6, # 2 per cloud provider
"complexity": "TIER_3",
"validates": [
"multi_entity_research",
"technical_comparison",
"synthesis_quality",
],
},
"test_5_edge_case": {
"name": "Edge Case (Requires Scrolling/Search)",
"query": (
"Find the exact date and location of Anthropic's Series C funding announcement. "
"Include the lead investors and round size."
),
"expected_sources": 2,
"complexity": "TIER_1",
"validates": [
"specific_fact_retrieval",
"search_web_page_usage",
"detail_accuracy",
],
},
# BONUS: The original failing test for regression checking
"test_0_baseline": {
"name": "Baseline (Original Bug Test)",
"query": (
"Find the 2024 revenue of NVIDIA and compare it to AMD's 2024 revenue. "
"Save your findings to the scratchpad before answering."
),
"expected_sources": 2, # 1 per company minimum
"complexity": "TIER_1",
"validates": [
"both_entities_researched",
"scratchpad_usage",
"basic_comparison",
],
},
}
PROMPTS = {
"RESEARCH": (
"Find the 2024 revenue of NVIDIA and compare it to AMD's 2024 revenue. "
"Save your findings to the scratchpad before answering."
),
"EXTERNAL": ("Check the flight times from LAX to JFK."),
}
config = {
"together_api_key": "",
"entities_api_key": "",
"entities_user_id": "",
"base_url": "http://localhost:9000",
"url": "https://api.together.xyz/v1/chat/completions",
"model": "together-ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"provider": "together",
"assistant_id": "asst_13HyDgBnZxVwh5XexYu74F",
"test_prompt": DEEP_RESEARCH_TEST_SUITE["test_0_baseline"][
"query"
], # The baseline/original test
# or
# "test_prompt": DEEP_RESEARCH_TEST_SUITE["test_1_deep_comparative"]["query"],
# or
# "test_prompt": DEEP_RESEARCH_TEST_SUITE["test_3_multi_source_verification"]["query"],
}