Phase 5: re-quantize merged checkpoint to NVFP4 with MTP/vision tensor reinjection and production-config verification #4

Merged
aleleba merged 16 commits from agente-fase5-quantize-nvfp4 into master 2026-07-30 06:41:14 -06:00
3 changed files with 3 additions and 3 deletions
Showing only changes of commit c9878ef98f - Show all commits
+1 -1
View File
@@ -32,7 +32,7 @@ import requests
REPO_ROOT = Path(__file__).resolve().parent.parent REPO_ROOT = Path(__file__).resolve().parent.parent
HOLDOUT_PATH = REPO_ROOT / "data" / "holdout_prompts.jsonl" HOLDOUT_PATH = REPO_ROOT / "data" / "holdout_prompts.jsonl"
RESULTS_PATH = REPO_ROOT / "data" / "gate2_results.json" RESULTS_PATH = REPO_ROOT / "data" / os.environ.get("GATE2_RESULTS_FILENAME", "gate2_results.json")
BASE_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001") BASE_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001")
MODEL_NAME = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16") MODEL_NAME = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16")
+1 -1
View File
@@ -27,7 +27,7 @@ from pathlib import Path
import requests import requests
REPO_ROOT = Path(__file__).resolve().parent.parent REPO_ROOT = Path(__file__).resolve().parent.parent
RESULTS_PATH = REPO_ROOT / "data" / "gate3_results.json" RESULTS_PATH = REPO_ROOT / "data" / os.environ.get("GATE3_RESULTS_FILENAME", "gate3_results.json")
EVAL_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001") EVAL_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001")
EVAL_MODEL = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16") EVAL_MODEL = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16")
+1 -1
View File
@@ -32,7 +32,7 @@ import requests
REPO_ROOT = Path(__file__).resolve().parent.parent REPO_ROOT = Path(__file__).resolve().parent.parent
SCHEMAS_DIR = REPO_ROOT / "data" / "schemas" SCHEMAS_DIR = REPO_ROOT / "data" / "schemas"
RESULTS_PATH = REPO_ROOT / "data" / "gate4_results.json" RESULTS_PATH = REPO_ROOT / "data" / os.environ.get("GATE4_RESULTS_FILENAME", "gate4_results.json")
EVAL_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001") EVAL_URL = os.environ.get("VLLM_EVAL_URL", "http://localhost:8001")
EVAL_MODEL = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16") EVAL_MODEL = os.environ.get("VLLM_EVAL_MODEL", "qwen3.6-35b-a3b-mcp-bf16")