config.yaml

schema: saqs.pack_install/v2
pack: Qwen3-Coder-Next-Spark-Agentic
release_class: day0_preview
clock_status: BLOCKED_until_2026-08-02_freeze_attestation
candidate_status: READY_FOR_FREEZE_REVIEW
source_candidate: true
verified_runnable_day0: false

pack_source:
  repository: https://github.com/hizrianraz/Qwen3-Coder-Next-Spark-Agentic.git
  revision: null
  revision_state: set_from_authorized_40_hex_release_commit_after_publish
  runtime_requires_external_commit_pin: QWEN_EXPECT_PACK_REVISION
  runtime_requires_canonical_root_pin: QWEN_ALLOWED_PACK_ROOT
  dirty_tracked_tree_allowed: false
  critical_runtime_paths_must_be_tracked: true

artifact:
  repository: Qwen/Qwen3-Coder-Next-FP8
  revision: da6e2ed27304dd39abadd9c82ef50e8de67bdd4c
  format: fp8_safetensors
  expected_shards: 40
  verified_files: 52
  checksum_manifest: results/SHA256SUMS.fp8
  checksum_manifest_sha256: d7269c078ec614ae57792ec63da00b3e13cf88e61f421ef4bcf4bdaac54b0190
  weights_in_pack: false
  diy_gguf: false
  downloader:
    executable_must_be_absolute_and_sha256_bound: true
    environment: clean_environment_private_ephemeral_home_no_operator_credentials
    destination: new_private_staging_then_no_overwrite_publish

upstream_profile:
  total_parameters_billions: 80
  active_parameters_per_token_billions: 3
  native_context_tokens: 262144
  launch_profile_context_tokens: 32768
  launch_profile_effective_context_measured: false

runtime:
  engine: vllm
  version: 0.25.1
  commit: 752a3a504485790a2e8491cacbb35c137339ad34
  image: vllm/vllm-openai@sha256:e4f88a835143cd22aee2397a26ec6bb80b3a4a6fe0c882bcbc63822904766089
  oci_index_digest: sha256:e4f88a835143cd22aee2397a26ec6bb80b3a4a6fe0c882bcbc63822904766089
  platform: linux/arm64
  platform_manifest_digest: sha256:2cc49b81319f7a66a33dd8bd63a7bfddae079122b33ce51989b6828a1f038c37
  image_config_digest: sha256:30a38a1d74a17365eca400e83ffd885b250e0c8c0d3c5b508afa8c412d2ddf95
  image_pull_policy_at_launch: never
  docker_context: default
  docker_endpoint: unix:///var/run/docker.sock
  docker_operations_force_explicit_local_endpoint: true
  docker_client_environment: clean
  remote_docker_allowed: false
  pull_script: scripts/pull_official_fp8.sh
  serve_script: scripts/serve_vllm_fp8.sh
  compatibility_serve_scripts:
    - scripts/serve_spark.sh
    - scripts/serve_spark_vllm.sh
  model_alias: local-qwen3-coder-next
  native_tool_parser: qwen3_coder
  trust_remote_code: false
  fastapi_docs_enabled: false

spark_profile:
  tensor_parallel_size: 1
  max_model_len: 32768
  max_num_seqs: 1
  gpu_memory_utilization: 0.80
  min_available_memory_gib_before_load: 96
  max_peak_nonswapped_gib_for_future_certification: 112
  min_free_memory_gib_after_load_for_future_certification: 12
  single_model_residency_required: true
  full_gpu_compute_process_inventory_required_at_launch: true
  post_lock_complete_snapshot_rehash_required: true
  post_lock_docker_boundary_revalidation_required: true
  freeze_operator_quiescence_attestation_required: true

locks:
  qwen_pull_serve_serialization: /run/user/<uid>/saqs-qwen3-coder-next-model-operation.lock.d
  shared_host_model_residency: /run/lock/saqs-dgx-spark-model-residency.lock.d
  stale_lock_policy: inspect_then_remove_manually

endpoint:
  protocol: openai_compatible
  base_url: http://127.0.0.1:8001/v1
  host_publication: 127.0.0.1
  port: 8001
  v1_authentication: VLLM_API_KEY_required
  vllm_api_key_path_prefixes: [/v1, /v2, /inference]
  other_paths_authenticated_by_vllm_api_key: false
  whole_server_authentication: false
  security_limit: vllm_api_key_only_covers_v1_v2_and_inference_prefixes
  deployment_trust_boundary: trusted_single_user_loopback_only
  docker_environment_is_secret_store: false
  loopback_limit: local_processes_and_users_can_reach_host_loopback
  remote_access: authenticated_ssh_tunnel_or_full_service_tls_auth_proxy

sampling:
  temperature: 0.0
  top_p: 1.0
  max_tokens: 512

claims:
  current_evidence: api_chat_availability_smoke_2_of_2_historical_only
  not_established:
    - exact_pinned_runtime_model_load
    - tool_call_correctness
    - closed_loop_agent_reliability
    - throughput_or_concurrency
    - saqs_certification
    - ship_or_hero_status