{ "schema_version": "ta-amps.partner-compute-cascade.v1", "mode": "local_verified", "credential_environment_status": { "MODAL_TOKEN_ID": "unset", "BASETEN_API_KEY": "unset", "FIREWORKS_API_KEY": "unset", "META_API_KEY": "unset" }, "honesty_statement": "All partner API keys are unset in this build environment (or live calls were not invoked); every receipt below is deterministically executed and verified locally (mode=local_verified), never faking live cloud API responses.", "modal_qlora_and_sandbox": { "provider": "modal", "mode": "local_verified", "live_credentials_present": false, "modal_sdk_installed": false, "honesty_note": "MODAL_TOKEN_ID is not set in this local environment; dataset formatting, 4-gram leakage gating, QLoRA NF4 VRAM math and container sandbox isolation contracts were verified deterministically offline.", "runnable_cloud_commands": [ "modal run packs/partner-compute-cascade/code/partner_cascade/modal_qlora_and_sandbox.py::sandbox_verify_pack --pack-id meta-are-scenario-matrix", "modal run packs/partner-compute-cascade/code/partner_cascade/modal_qlora_and_sandbox.py::train_qlora --model-name Muse-Glimmer-30B", "modal deploy packs/partner-compute-cascade/code/partner_cascade/modal_qlora_and_sandbox.py" ], "dataset_validation": { "accepted_count": 1, "rejected_count": 1, "dataset_sha256": "25bd2319feb702450cc1eddabeae477ad726a2bf0ee7abecc45d3166add0fabd" }, "qlora_configs": { "Llama-3.2-3B-Instruct": { "model_name": "Llama-3.2-3B-Instruct", "hf_repo": "meta-llama/Llama-3.2-3B-Instruct", "quantization": "nf4_double_quant_bf16_compute", "lora_rank": 16, "lora_alpha": 32, "lora_scaling": 2.0, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj" ], "trainable_params": 24313856, "total_params": 3212749824, "trainable_pct": 0.7568, "base_nf4_gib": 1.646, "lora_and_optimizer_gib": 0.136, "activation_and_cuda_gib": 2.306, "estimated_peak_vram_gib": 4.088, "recommended_modal_gpu": "L4", "gpu_vram_gib": 24.0, "fits_within_92pct_vram": true, "modal_hourly_usd": 0.8, "runs_per_100_usd_credits_20min_job": 375 }, "Muse-Glimmer-30B": { "model_name": "Muse-Glimmer-30B", "hf_repo": "meta-muse/Muse-Glimmer-30B", "quantization": "nf4_double_quant_bf16_compute", "lora_rank": 32, "lora_alpha": 64, "lora_scaling": 2.0, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj" ], "trainable_params": 204472320, "total_params": 30450000000, "trainable_pct": 0.6715, "base_nf4_gib": 15.597, "lora_and_optimizer_gib": 1.143, "activation_and_cuda_gib": 2.962, "estimated_peak_vram_gib": 19.702, "recommended_modal_gpu": "A100-40GB", "gpu_vram_gib": 40.0, "fits_within_92pct_vram": true, "modal_hourly_usd": 2.5, "runs_per_100_usd_credits_20min_job": 120 } }, "sandbox_contract": { "contract_valid": true, "violations": [], "sandbox_spec_sha256": "64f841657554c46a6f201796790fb365804e3ca96d69de2469f5e906ec00f8c2" } }, "baseten_2m_tpm_swarm": { "provider": "baseten", "mode": "local_verified", "live_credentials_present": false, "max_tpm_ceiling": 2000000, "peak_sliding_window_tokens": 1980000, "respected_2m_tpm_ceiling": true, "total_claims_submitted": 14, "succeeded_claims": 12, "pre_flight_fenced_or_ttl_blocked": 2, "retried_429_preserved_in_ledger": true, "pacing_waits_triggered": 1, "virtual_seconds_paced": 60.001, "total_tokens_processed": 2160000, "receipt_sha256": "0f70ff7ab8ecac48d8faec1a54457d8881fe470c468b7d8402400211ccf0a943" }, "fireworks_multilora_shim": { "provider": "fireworks_ai", "mode": "local_verified", "live_credentials_present": false, "pattern_attribution": "Sinan Ozdemir specialist multi-LoRA routing and explicit rubric verification behind <2ms edge shim", "specialist_adapter_count": 4, "initial_route": { "tier": "fireworks_specialist_multilora", "fingerprint": "4e038dc44ca72a7cd79b69d7adbae7228b8dd53c5ccb7200b0e7a6bdf597870d", "selected_adapter": "statute-citation-extractor-lora", "fireworks_model_id": "accounts/ta-amps/models/llama3p1-8b-statute-citation", "verifier_adapter": "rubric-judge-verifier-lora", "response_format": { "type": "json_schema", "json_schema": { "name": "tessera_task_output", "strict": true, "required": [ "summary", "citations", "human_only" ] } }, "shim_latency_ms": 0.03924, "within_2ms_shim_budget": true }, "memo_replay_tier": "edge_shim_zero_token", "shim_within_2ms": true, "good_candidate_passed": true, "self_verification_blocked": true, "pii_leak_escalated": true, "receipt_sha256": "9b8ad34a6f1e5fe2810bdd098d0d61feffdae75c2f0556b6ad9d00d8a6c9713a" }, "cascade_1000_request_simulation": { "workload_requests": 1000, "tiers": { "tier_0_edge_shim_memo": { "requests": 310, "avg_tokens": 0, "avg_latency_ms": 0.08, "cost_per_request_usd": 0.0 }, "tier_1_fireworks_multilora": { "requests": 410, "avg_tokens": 2400, "avg_latency_ms": 420.0, "cost_per_request_usd": 0.00102 }, "tier_2_baseten_2m_tpm_swarm": { "requests": 180, "avg_tokens": 6200, "avg_latency_ms": 1650.0, "cost_per_request_usd": 0.0038 }, "tier_3_modal_qlora_and_sandbox": { "requests": 70, "avg_tokens": 4800, "avg_latency_ms": 2850.0, "cost_per_request_usd": 0.0064 }, "tier_4_meta_muse_spark_frontier": { "requests": 30, "avg_tokens": 5500, "avg_latency_ms": 3900.0, "cost_per_request_usd": 0.0095 } }, "cascade_total_cost_usd": 1.8352, "naive_all_frontier_cost_usd": 9.5, "cost_savings_pct": 80.68, "weighted_avg_latency_ms": 785.72, "partner_credit_envelopes_300_usd": { "modal_100_usd": { "credit_usd": 100.0, "llama3_3b_qlora_l4_20min_runs": 375, "muse_glimmer_30b_qlora_a100_20min_runs": 120, "isolated_pack_sandbox_verifications": 45000 }, "baseten_100_usd": { "credit_usd": 100.0, "rate_ceiling_tpm": 2000000, "sustained_hourly_token_throughput": 120000000, "estimated_public_statute_swarm_tasks": 26300 }, "fireworks_100_usd": { "credit_usd": 100.0, "8b_multilora_tokens_supported": 500000000, "rubric_verified_triage_requests": 98000 } } }, "artifact_sha256": "1c37b813780cb6ce946cad0061b234634bd2088645557947fb343f7efbaa6f5e" }