hud-python 0.4.40__tar.gz → 0.4.42__tar.gz

{hud_python-0.4.40 → hud_python-0.4.42}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.40
+Version: 0.4.42
 Summary: SDK for the HUD platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-python
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues
@@ -159,10 +159,10 @@ OSS RL environment + evals toolkit. Wrap software as environments, run benchmark
 
 ## Highlights
 
+- 🎓 **[One-click RL](https://hud.so/models)** – Run `hud rl` to get a trained model on any environment.
 - 🚀 **[MCP environment skeleton](https://docs.hud.so/core-concepts/mcp-protocol)** – any agent can call any environment.
 - ⚡️ **[Live telemetry](https://hud.so)** – inspect every tool call, observation, and reward in real time.
 - 🗂️ **[Public benchmarks](https://hud.so/leaderboards)** – OSWorld-Verified, SheetBench-50, and more.
-- 🌱 **[Reinforcement learning built-in](rl/)** – Verifiers gym pipelines for GRPO on any environment.
 - 🌐 **[Cloud browsers](environments/remote_browser/)** – AnchorBrowser, Steel, BrowserBase integrations for browser automation.
 - 🛠️ **[Hot-reload dev loop](environments/README.md#phase-5-hot-reload-development-with-cursor-agent)** – `hud dev` for iterating on environments without rebuilds.
 
@@ -171,27 +171,46 @@ OSS RL environment + evals toolkit. Wrap software as environments, run benchmark
 ## Installation
 
 ```bash
-# Core installation - MCP servers, telemetry, basic tools for environment design
+# SDK - MCP servers, telemetry, evaluation
 pip install hud-python
 
-# Agent installation - Adds AI providers, datasets
-pip install "hud-python[agent]"
-
-# CLI utilities
+# CLI - RL pipeline, environment design
 uv tool install hud-python
 # uv tool update-shell
-
-# From source (latest)
-git clone https://github.com/hud-evals/hud-python
-pip install -e "hud-python[dev]"
 ```
 
 > See [docs.hud.so](https://docs.hud.so), or add docs to any MCP client:
 > `claude mcp add --transport http docs-hud https://docs.hud.so/mcp`
 
-## Quickstart
+Before starting, get your HUD_API_KEY at [hud.so](https://hud.so).
+
+
+## Quickstart: Training
+
+RL using GRPO a Qwen2.5-VL model on any hud dataset:
+
+```bash
+hud get hud-evals/basic-2048 # from HF
+hud rl basic-2048.json
+```
+
+> See [agent training docs](https://docs.hud.so/train-agents/quickstart)
 
-For a tutorial that explains the agent and evaluation design, run ([see quickstart docs](https://docs.hud.so/quickstart)):
+Or make your own environment and dataset:
+
+```bash
+hud init my-env && cd my-env
+hud dev --interactive
+# When ready to run:
+hud rl
+```
+
+> See [environment design docs](https://docs.hud.so/build-environments)
+
+
+## Quickstart: Evals
+
+For a tutorial that explains the agent and evaluation design, run:
 
 ```python
 uvx hud-python quickstart
@@ -262,20 +281,22 @@ hud rl hud-evals/basic-2048
 
 # Option B: Download first, modify, then train
 hud get hud-evals/basic-2048
-hud rl basic-2048.jsonl
+hud rl basic-2048.json
 
 # Optional: baseline evaluation
-hud eval basic-2048.jsonl
+hud eval basic-2048.json
 ```
 
 Supports multi‑turn RL for both:
 - Language‑only models (e.g., `Qwen/Qwen2.5-7B-Instruct`)
 - Vision‑Language models (e.g., `Qwen/Qwen2.5-VL-3B-Instruct`)
 
-By default, `hud rl` provisions a persistant server and trainer in the cloud, streams telemetry to `hud.so`, and lets you monitor/manage models at `hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
+By default, `hud rl` provisions a persistent server and trainer in the cloud, streams telemetry to `hud.so`, and lets you monitor/manage models at `hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
 
 Any HUD MCP environment and evaluation works with our RL pipeline (including remote configurations). See the guided docs: `https://docs.hud.so/train-agents/quickstart`.
 
+Pricing: Hosted vLLM and training GPU rates are listed in the [Training Quickstart → Pricing](https://docs.hud.so/train-agents/quickstart#pricing). Manage billing at the [HUD billing dashboard](https://hud.so/project/billing).
+
 ## Benchmarking Agents
 
 This is Claude Computer Use running on our proprietary financial analyst benchmark [SheetBench-50](https://huggingface.co/datasets/hud-evals/SheetBench-50):
@@ -323,7 +344,7 @@ from hud.tools import HudComputerTool
 mcp = MCPServer("My Environment")
 
 # Add hud tools (see all tools: https://docs.hud.so/reference/tools)
-mcp.add_tool(HudComputerTool())
+mcp.tool(HudComputerTool())
 
 # Or custom tools (see https://docs.hud.so/build-environments/adapting-software)
 @mcp.tool("launch_app"):
@@ -494,11 +515,10 @@ graph LR
 
 ## Roadmap
 
-- Merging our forks in to the main `mcp`, `mcp_use`, `verifiers` repositories
+- Merging our forks in to the main `mcp`, `mcp_use` repositories
 - Helpers for building new environments (see [current guide](environments/README.md))
 - Integrations with every major agent framework
 - Evaluation environment registry
-- Native RL training to hud environments (see [current RL support](rl/))
 - MCP opentelemetry standard
 
 ## Contributing
@@ -509,7 +529,7 @@ Key areas:
 - [Environment examples](environments/) - Add new MCP environments
 - [Agent implementations](hud/agents/) - Add support for new LLM providers
 - [Tool library](hud/tools/) - Extend the built-in tool collection
-- [RL training](rl/) - Improve reinforcement learning pipelines
+- [RL training](hud/rl/) - Improve reinforcement learning pipelines
 
 Thanks to all our contributors!
 

{hud_python-0.4.40 → hud_python-0.4.42}/README.md

@@ -22,10 +22,10 @@ OSS RL environment + evals toolkit. Wrap software as environments, run benchmark
 
 ## Highlights
 
+- 🎓 **[One-click RL](https://hud.so/models)** – Run `hud rl` to get a trained model on any environment.
 - 🚀 **[MCP environment skeleton](https://docs.hud.so/core-concepts/mcp-protocol)** – any agent can call any environment.
 - ⚡️ **[Live telemetry](https://hud.so)** – inspect every tool call, observation, and reward in real time.
 - 🗂️ **[Public benchmarks](https://hud.so/leaderboards)** – OSWorld-Verified, SheetBench-50, and more.
-- 🌱 **[Reinforcement learning built-in](rl/)** – Verifiers gym pipelines for GRPO on any environment.
 - 🌐 **[Cloud browsers](environments/remote_browser/)** – AnchorBrowser, Steel, BrowserBase integrations for browser automation.
 - 🛠️ **[Hot-reload dev loop](environments/README.md#phase-5-hot-reload-development-with-cursor-agent)** – `hud dev` for iterating on environments without rebuilds.
 
@@ -34,27 +34,46 @@ OSS RL environment + evals toolkit. Wrap software as environments, run benchmark
 ## Installation
 
 ```bash
-# Core installation - MCP servers, telemetry, basic tools for environment design
+# SDK - MCP servers, telemetry, evaluation
 pip install hud-python
 
-# Agent installation - Adds AI providers, datasets
-pip install "hud-python[agent]"
-
-# CLI utilities
+# CLI - RL pipeline, environment design
 uv tool install hud-python
 # uv tool update-shell
-
-# From source (latest)
-git clone https://github.com/hud-evals/hud-python
-pip install -e "hud-python[dev]"
 ```
 
 > See [docs.hud.so](https://docs.hud.so), or add docs to any MCP client:
 > `claude mcp add --transport http docs-hud https://docs.hud.so/mcp`
 
-## Quickstart
+Before starting, get your HUD_API_KEY at [hud.so](https://hud.so).
+
+
+## Quickstart: Training
+
+RL using GRPO a Qwen2.5-VL model on any hud dataset:
+
+```bash
+hud get hud-evals/basic-2048 # from HF
+hud rl basic-2048.json
+```
+
+> See [agent training docs](https://docs.hud.so/train-agents/quickstart)
 
-For a tutorial that explains the agent and evaluation design, run ([see quickstart docs](https://docs.hud.so/quickstart)):
+Or make your own environment and dataset:
+
+```bash
+hud init my-env && cd my-env
+hud dev --interactive
+# When ready to run:
+hud rl
+```
+
+> See [environment design docs](https://docs.hud.so/build-environments)
+
+
+## Quickstart: Evals
+
+For a tutorial that explains the agent and evaluation design, run:
 
 ```python
 uvx hud-python quickstart
@@ -125,20 +144,22 @@ hud rl hud-evals/basic-2048
 
 # Option B: Download first, modify, then train
 hud get hud-evals/basic-2048
-hud rl basic-2048.jsonl
+hud rl basic-2048.json
 
 # Optional: baseline evaluation
-hud eval basic-2048.jsonl
+hud eval basic-2048.json
 ```
 
 Supports multi‑turn RL for both:
 - Language‑only models (e.g., `Qwen/Qwen2.5-7B-Instruct`)
 - Vision‑Language models (e.g., `Qwen/Qwen2.5-VL-3B-Instruct`)
 
-By default, `hud rl` provisions a persistant server and trainer in the cloud, streams telemetry to `hud.so`, and lets you monitor/manage models at `hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
+By default, `hud rl` provisions a persistent server and trainer in the cloud, streams telemetry to `hud.so`, and lets you monitor/manage models at `hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
 
 Any HUD MCP environment and evaluation works with our RL pipeline (including remote configurations). See the guided docs: `https://docs.hud.so/train-agents/quickstart`.
 
+Pricing: Hosted vLLM and training GPU rates are listed in the [Training Quickstart → Pricing](https://docs.hud.so/train-agents/quickstart#pricing). Manage billing at the [HUD billing dashboard](https://hud.so/project/billing).
+
 ## Benchmarking Agents
 
 This is Claude Computer Use running on our proprietary financial analyst benchmark [SheetBench-50](https://huggingface.co/datasets/hud-evals/SheetBench-50):
@@ -186,7 +207,7 @@ from hud.tools import HudComputerTool
 mcp = MCPServer("My Environment")
 
 # Add hud tools (see all tools: https://docs.hud.so/reference/tools)
-mcp.add_tool(HudComputerTool())
+mcp.tool(HudComputerTool())
 
 # Or custom tools (see https://docs.hud.so/build-environments/adapting-software)
 @mcp.tool("launch_app"):
@@ -357,11 +378,10 @@ graph LR
 
 ## Roadmap
 
-- Merging our forks in to the main `mcp`, `mcp_use`, `verifiers` repositories
+- Merging our forks in to the main `mcp`, `mcp_use` repositories
 - Helpers for building new environments (see [current guide](environments/README.md))
 - Integrations with every major agent framework
 - Evaluation environment registry
-- Native RL training to hud environments (see [current RL support](rl/))
 - MCP opentelemetry standard
 
 ## Contributing
@@ -372,7 +392,7 @@ Key areas:
 - [Environment examples](environments/) - Add new MCP environments
 - [Agent implementations](hud/agents/) - Add support for new LLM providers
 - [Tool library](hud/tools/) - Extend the built-in tool collection
-- [RL training](rl/) - Improve reinforcement learning pipelines
+- [RL training](hud/rl/) - Improve reinforcement learning pipelines
 
 Thanks to all our contributors!
 

{hud_python-0.4.40 → hud_python-0.4.42}/environments/blank/pyproject.toml

@@ -3,7 +3,7 @@ name = "test_test"
 version = "0.1.0"
 description = "A minimal HUD environment"
 requires-python = ">=3.11"
-dependencies = [ "hud-python==0.4.40", "fastapi", "uvicorn[standard]", "httpx>=0.28.1",]
+dependencies = [ "hud-python==0.4.41", "fastapi", "uvicorn[standard]", "httpx>=0.28.1",]
 
 [build-system]
 requires = [ "hatchling",]

{hud_python-0.4.40 → hud_python-0.4.42}/environments/deepresearch/pyproject.toml

@@ -3,7 +3,7 @@ name = "deepresearch"
 version = "0.1.0"
 description = "DeepResearch HUD environment with HTTP backend (EXA on server)"
 requires-python = ">=3.11"
-dependencies = [ "hud-python==0.4.40", "fastapi>=0.104.1", "uvicorn[standard]>=0.24.0", "httpx>=0.24.0",]
+dependencies = [ "hud-python==0.4.41", "fastapi>=0.104.1", "uvicorn[standard]>=0.24.0", "httpx>=0.24.0",]
 
 [build-system]
 requires = [ "hatchling",]

{hud_python-0.4.40 → hud_python-0.4.42}/hud/agents/__init__.py

@@ -2,14 +2,12 @@ from __future__ import annotations
 
 from .base import MCPAgent
 from .claude import ClaudeAgent
-from .lite_llm import LiteAgent
 from .openai import OperatorAgent
 from .openai_chat_generic import GenericOpenAIChatAgent
 
 __all__ = [
     "ClaudeAgent",
     "GenericOpenAIChatAgent",
-    "LiteAgent",
     "MCPAgent",
     "OperatorAgent",
 ]

{hud_python-0.4.40 → hud_python-0.4.42}/hud/agents/openai_chat_generic.py

@@ -204,7 +204,9 @@ class GenericOpenAIChatAgent(MCPAgent):
 
         try:
             response = await self._invoke_chat_completion(
-                messages=messages, tools=tools, extra=extra # type: ignore
+                messages=messages,
+                tools=tools, # type: ignore
+                extra=extra,
             )
         except Exception as e:
             error_content = f"Error getting response {e}"

{hud_python-0.4.40 → hud_python-0.4.42}/hud/cli/flows/tasks.py

@@ -212,14 +212,14 @@ def convert_tasks_to_remote(tasks_file: str) -> str:
     # Check if tasks already have remote URLs
     already_remote = _validate_tasks(tasks)
 
+    # If tasks already reference a remote MCP URL, do not require a local environment
+    # or attempt any image updates. Use the dataset as-is.
+    if already_remote:
+        return str(tasks_path)
+
     # Extract existing images from tasks
     existing_images = _extract_existing_images(tasks)
 
-    # Load tasks (supports .json and .jsonl)
-    if already_remote and not existing_images:
-        # Tasks are remote but have no image references - just return as-is
-        return str(tasks_path)
-
     # Locate environment
     env_dir = find_environment_dir(tasks_path)
     if not env_dir:

{hud_python-0.4.40 → hud_python-0.4.42}/hud/cli/rl/gpu_utils.py

@@ -253,6 +253,8 @@ def adjust_config_for_ddp(config: Config, num_gpus: int) -> Config:
     # Update max_parallel_episodes to match
     config.actor.max_parallel_episodes = config.training.batch_size
 
+    config.training.num_gpus = num_gpus
+
     # Log the adjustment
     from rich.console import Console
 

{hud_python-0.4.40 → hud_python-0.4.42}/hud/cli/rl/remote_runner.py

@@ -13,6 +13,7 @@ from pathlib import Path
 from rich.console import Console
 
 from hud.cli.rl.celebrate import show_confetti_async
+from hud.cli.rl.gpu_utils import adjust_config_for_ddp
 from hud.cli.rl.viewer import show_json_interactive
 from hud.cli.rl.wait_utils import wait_for_enter_cancel_or_change
 from hud.utils.hud_console import hud_console
@@ -309,7 +310,7 @@ def run_remote_training(
         # console.print(gpu_table)
 
         if yes:
-            gpu_choice = "A100"  # Default GPU in yes mode
+            gpu_choice = "A100"
             hud_console.info(f"Auto-selecting GPU: {gpu_choice} 80GB (--yes mode)")
         else:
             gpu_choice = hud_console.select(
@@ -322,7 +323,7 @@ def run_remote_training(
             )
 
         if yes:
-            num_gpus = 1  # Default to 1 GPU in yes mode
+            num_gpus = 2 # Default to 2 GPUs in yes mode
             hud_console.info(f"Auto-selecting {num_gpus} GPU(s) (--yes mode)")
         else:
             num_gpus = hud_console.select(
@@ -347,6 +348,10 @@ def run_remote_training(
             yes=yes,
         )
 
+        config = adjust_config_for_ddp(config, int(num_gpus))
+
+        config.training.gpu_type = gpu_choice
+
         # Use a short label for tasks (avoid full absolute paths)
         try:
             if tasks_file and Path(tasks_file).exists():
@@ -357,7 +362,7 @@ def run_remote_training(
         except Exception:
             tasks_label = str(tasks_file)
 
-        config.job_name = f"RL {model_name} on {tasks_label}"
+        config.job_name = f"RL {tasks_label} | {model_name}"
 
         # Save config so user can review/edit externally
         temp_config_path = Path(f".rl_config_temp_{model_name}.json")
@@ -421,8 +426,8 @@ def run_remote_training(
         hud_console.info(f"Loading configuration from: {config_file}")
         config = load_config(config_file)
         config_dict = config.to_dict()
-        gpu_choice = "A100"  # Default
-        num_gpus = 1  # Default for non-interactive mode
+        gpu_choice = config.training.gpu_type
+        num_gpus = config.training.num_gpus
 
     # Launch training
     try:

{hud_python-0.4.40 → hud_python-0.4.42}/hud/rl/actor.py

@@ -109,7 +109,7 @@ class Actor:
 
         # Run the task
         try:
-            with hud.trace(f"Training | {task.id}", job_id=job_id):
+            with hud.trace(f"Training | {task.prompt}", job_id=job_id):
                 result = await agent.run(task, max_steps=self.actor_config.max_steps_per_episode)
 
         except Exception:

{hud_python-0.4.40 → hud_python-0.4.42}/hud/rl/config.py

@@ -61,6 +61,9 @@ class ModelConfig:
 @dataclass
 class TrainingConfig:
     """Training hyperparameters."""
+    # GPU parameters
+    gpu_type: str = "A100"
+    num_gpus: int = 2
 
     # Training parameters
     training_steps: int = 100

{hud_python-0.4.40 → hud_python-0.4.42}/hud/rl/train.py

@@ -103,7 +103,10 @@ async def train(config: Config, tasks: list[Task]) -> None:
     if is_main_process():
         hud_console.info(f"Creating job with config.job_id: {config.job_id}")
         job_obj = hud.create_job(
-            job_id=config.job_id, name=config.job_name, metadata={"config": config.to_dict()}
+            job_id=config.job_id, name=config.job_name, metadata={
+                "config": config.to_dict(),
+                "agent_class": config.model.base_model
+            }
         )
         hud_console.info(f"Created job with job_obj.id: {job_obj.id}")
         job_obj.update_status_sync("running")

{hud_python-0.4.40 → hud_python-0.4.42}/hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.4.40"
+    assert hud.__version__ == "0.4.42"

{hud_python-0.4.40 → hud_python-0.4.42}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.40"
+__version__ = "0.4.42"

{hud_python-0.4.40 → hud_python-0.4.42}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.4.40"
+version = "0.4.42"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.13"