Upload folder using huggingface_hub

README.md

@@ -424,12 +424,12 @@ curl http://localhost:8000/state
 
 Scores generated by running `uv run inference.py` with `google/gemma-4-31B-it` via the HuggingFace Router.
 
-| Task | Difficulty | Score | Steps | Result | Notes |
-| ---- | ---------- | ----- | ----- | ------ | ----- |
-| `easy_list_pants` | Easy | **0.74** | 6 | PASS | List products in 'Pants' category |
-| `medium_cart_camera_backpack` | Medium | **0.56** | 20 | PASS | Add Camera Backpack to guest cart |
-| `medium_cart_flannel_jacket` | Medium | **0.60** | 20 | PASS | Add Flannel Jacket to guest cart |
-| `hard_checkout_ripstop_pants` | Hard | **0.22** | 20 | FAIL | Full guest checkout (hit step limit) |
-| **Overall** | — | **0.53** | — | **3/4 passed** | |
+| Task | Difficulty | Score | Steps | Notes |
+| ---- | ---------- | ----- | ----- | ----- |
+| `easy_list_pants` | Easy | **0.74** | 7 | List products in 'Pants' category |
+| `medium_cart_camera_backpack` | Medium | **0.46** | 20 | Add Camera Backpack to guest cart |
+| `medium_cart_flannel_jacket` | Medium | **0.52** | 20 | Add Flannel Jacket to guest cart |
+| `hard_checkout_ripstop_pants` | Hard | **0.14** | 20 | Full guest checkout (hit step limit) |
+| **Overall** | — | **0.47** | — | |
 
 > **To regenerate:** `HF_TOKEN=hf_xxx uv run inference.py`

hars/shopping.har

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8bf3049de2939281f6e79cc61eef56665b1f506b4e14713e8cf730c6d78c6f77
-size 134
+oid sha256:878c65126d999ef91d6b75438431f7c1b9164ac580140bd7ca61ef693cacd76c
+size 115555293

hars/shopping_admin.har

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ced8b7d5612ce64cf669ab8b43080de3b1ed4384b0d9064c12c21f2b7f5d4665
-size 134
+oid sha256:ce2209be9f3265b0a1682935171fb932c0056bc67b7517419b3ef5239c2ba2be
+size 148077790

inference.py

@@ -707,7 +707,7 @@ async def run_episode(task_config: dict, client: OpenAI) -> dict:
 
     rewards: List[float] = []
     steps_taken = 0
-    score = 0.0
+    score = 0.001  # strict (0, 1) required by validator; 0.0 / 1.0 are rejected
     success = False
     last_result = None
     history: List[dict] = []
@@ -790,7 +790,7 @@ async def run_episode(task_config: dict, client: OpenAI) -> dict:
         # Shift by +1.5 so that the fail reward (-1.5) maps to 0 and max maps to 1.
         _TEMPLATE_REWARD_CEIL = {1: 2.0, 3: 3.5, 6: 5.0}
         _reward_ceil = _TEMPLATE_REWARD_CEIL.get(task_config.get("template_id"), 5.0)
-        score = max(0.0, min(1.0, (total_reward + 1.5) / (_reward_ceil + 1.5)))
+        score = max(0.001, min(0.999, (total_reward + 1.5) / (_reward_ceil + 1.5)))
         success = total_reward >= 0.5   # any positive terminal reward = success
 
         vprint(f"\n[VERBOSE] ── episode end — {task_name} ──")

openenv_harvestgym.egg-info/SOURCES.txt

@@ -15,6 +15,7 @@ server/tools/__init__.py
 server/tools/browser_agent.py
 server/tools/curl_exec.py
 server/tools/embed_cache.py
+server/tools/html_distiller.py
 server/tools/search_endpoints.py
 server/tools/search_episode_data.py
 tests/test_e2e_episode.py