Spaces:
Sleeping
Sleeping
File size: 5,497 Bytes
ba71ff7 eca0e48 11cb018 ba71ff7 eca0e48 ba71ff7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 | # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
"""
FastAPI application for the REPL Environment.
This module creates an HTTP server that exposes the REPLEnvironment
over HTTP and WebSocket endpoints, compatible with EnvClient.
The server includes llm_query and llm_query_batched support via HuggingFace Inference API,
enabling the Recursive Language Model (RLM) paradigm.
LLM Token Configuration:
1. Client can pass `hf_token` in reset() - RECOMMENDED
2. Server fallback: HF_TOKEN environment variable
LLM functions are created dynamically in REPLEnvironment.reset() based on the
available token (client or server).
Usage:
# Development (with auto-reload):
uvicorn server.app:app --reload --host 0.0.0.0 --port 8000
# Production:
uvicorn server.app:app --host 0.0.0.0 --port 8000 --workers 4
# Or run directly:
uv run --project . server
Environment Variables:
HF_TOKEN: Fallback HuggingFace API token (client token takes priority)
LLM_MODEL: Model to use for llm_query/llm_query_batched (default: Qwen/Qwen3.5-9B)
MAX_CONCURRENT_ENVS: Maximum concurrent WebSocket sessions (default: 8)
"""
import inspect
import logging
import os
try:
from openenv.core.env_server.http_server import create_app
from ..models import REPLAction, REPLObservation
from .gradio_ui import build_repl_gradio_app
from .repl_environment import REPLEnvironment
except ImportError:
from models import REPLAction, REPLObservation
from openenv.core.env_server.http_server import create_app
from server.gradio_ui import build_repl_gradio_app
from server.repl_environment import REPLEnvironment
# ============== CONFIGURATION ==============
LLM_MODEL = os.environ.get("LLM_MODEL", "Qwen/Qwen3.5-9B")
HF_TOKEN = os.environ.get("HF_TOKEN")
REPL_MAX_ITERATIONS = int(os.environ.get("REPL_MAX_ITERATIONS", "30"))
REPL_MAX_OUTPUT_LENGTH = int(os.environ.get("REPL_MAX_OUTPUT_LENGTH", "8192"))
REPL_CONTEXT_PREVIEW_LENGTH = int(os.environ.get("REPL_CONTEXT_PREVIEW_LENGTH", "500"))
REPL_RLM_MAX_DEPTH = int(os.environ.get("REPL_RLM_MAX_DEPTH", "2"))
REPL_RLM_MAX_ITERATIONS = int(os.environ.get("REPL_RLM_MAX_ITERATIONS", "30"))
MAX_CONCURRENT_ENVS = int(os.environ.get("MAX_CONCURRENT_ENVS", "8"))
# ==========================================
_logger = logging.getLogger(__name__)
# Log LLM configuration
if HF_TOKEN:
print("[REPL Server] LLM support ENABLED (server token configured)")
print(f"[REPL Server] Default model: {LLM_MODEL}")
else:
print("[REPL Server] No server HF_TOKEN configured")
print(
"[REPL Server] LLM functions will be enabled if client passes hf_token in reset()"
)
def create_repl_environment() -> REPLEnvironment:
"""Factory function that creates REPLEnvironment with server config.
LLM functions are created dynamically during `reset()` when a client
passes `hf_token`. Rewards are computed via the default `REPLRubric`;
pass `expected_answer` at reset time for outcome-based scoring.
"""
return REPLEnvironment(
max_iterations=REPL_MAX_ITERATIONS,
max_output_length=REPL_MAX_OUTPUT_LENGTH,
context_preview_length=REPL_CONTEXT_PREVIEW_LENGTH,
rlm_max_depth=REPL_RLM_MAX_DEPTH,
rlm_max_iterations=REPL_RLM_MAX_ITERATIONS,
)
# Create the app with web interface and README integration.
_sig = inspect.signature(create_app)
if "gradio_builder" in _sig.parameters:
# Opt in to the primary-tab behaviour when supported so visitors land on
# the custom REPL UI instead of the auto-generated schema Playground.
# Older openenv-core versions silently ignore the extra kwargs.
create_app_kwargs: dict = {
"env_name": "repl_env",
"max_concurrent_envs": MAX_CONCURRENT_ENVS,
"gradio_builder": build_repl_gradio_app,
}
if "custom_tab_name" in _sig.parameters:
create_app_kwargs["custom_tab_name"] = "REPL"
if "custom_tab_primary" in _sig.parameters:
create_app_kwargs["custom_tab_primary"] = True
if "show_default_tab" in _sig.parameters:
# The auto-generated Playground is not useful for REPL (its real
# surface is injected Python helpers, not the bare action schema),
# so serve only the custom tab.
create_app_kwargs["show_default_tab"] = False
if "title_override" in _sig.parameters:
create_app_kwargs["title_override"] = (
"OpenEnv REPL — Recursive Language Model playground"
)
app = create_app(
create_repl_environment,
REPLAction,
REPLObservation,
**create_app_kwargs,
)
else:
_logger.warning(
"Installed openenv-core does not support gradio_builder; "
"custom REPL Gradio tab will not be available."
)
app = create_app(
create_repl_environment,
REPLAction,
REPLObservation,
env_name="repl_env",
max_concurrent_envs=MAX_CONCURRENT_ENVS,
)
def main():
"""
Entry point for direct execution via uv run or python -m.
This function enables running the server without Docker:
uv run --project . server
python -m envs.repl_env.server.app
openenv serve repl_env
"""
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
if __name__ == "__main__":
main()
|