Text Generation
Transformers
Safetensors
English
canopy
browser-use
web-agent
recurrent-moe
edge-llm
lightpanda
obscura
multi-agent
robotics-web
conversational
custom_code
Instructions to use psikosen/canopy-258m-r3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use psikosen/canopy-258m-r3 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="psikosen/canopy-258m-r3", trust_remote_code=True) messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("psikosen/canopy-258m-r3", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use psikosen/canopy-258m-r3 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "psikosen/canopy-258m-r3" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "psikosen/canopy-258m-r3", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/psikosen/canopy-258m-r3
- SGLang
How to use psikosen/canopy-258m-r3 with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "psikosen/canopy-258m-r3" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "psikosen/canopy-258m-r3", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "psikosen/canopy-258m-r3" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "psikosen/canopy-258m-r3", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use psikosen/canopy-258m-r3 with Docker Model Runner:
docker model run hf.co/psikosen/canopy-258m-r3
File size: 12,836 Bytes
ed79a7f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 | """
High-Performance Playwright Browser Controller.
Manages browser lifecycle, isolated contexts, stealth flags, fast execution,
and visual annotated screenshot auditing.
"""
from __future__ import annotations
import asyncio
import io
import logging
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple, Union
from PIL import Image, ImageDraw
from playwright.async_api import Browser, BrowserContext, Page, Error as PlaywrightError, async_playwright
from miniswardbower.browser.action_executor import ActionExecutor
from miniswardbower.browser.dom_distiller import DOMDistiller
from miniswardbower.core.config import BrowserConfig
from miniswardbower.core.schemas import ActionBatch, BrowserAction, BrowserActionType, PrunedAXTree
class BrowserController:
"""Async Playwright browser engine for the multi-agent team."""
def __init__(self, config: Optional[BrowserConfig] = None):
self.config = config or BrowserConfig()
self._playwright = None
self._browser: Optional[Browser] = None
self._context: Optional[BrowserContext] = None
self._page: Optional[Page] = None
self._lightpanda_driver = None
self._obscura_driver = None
self.distiller = DOMDistiller()
# Obscura 0.2.2 hit-testing lets pointer-events:none overlays intercept
# clicks. Keep annotations in the captured image until upstream fixes it.
if self.config.browser_engine == "obscura":
self.distiller.draw_overlays = False
self.executor = ActionExecutor(action_timeout_ms=self.config.action_timeout_ms, enable_natural_dynamics=True)
async def start(self) -> None:
"""Launches the configured browser; failures never silently change engines."""
if self._playwright is not None or self._lightpanda_driver is not None or self._obscura_driver is not None:
return
if self.config.browser_engine in ("lightpanda", "obscura"):
if self.config.browser_engine == "obscura":
from miniswardbower.browser.obscura_driver import ObscuraDriver
driver = ObscuraDriver(self.config.obscura_bin_path, self.config.obscura_allow_private_network)
else:
from miniswardbower.browser.lightpanda_driver import LightpandaDriver
driver = LightpandaDriver(bin_path=self.config.lightpanda_bin_path)
try:
self._browser, self._context, self._page = await driver.start()
await self.page.set_viewport_size({"width": self.config.window_width, "height": self.config.window_height})
except BaseException:
await driver.stop()
self._browser = self._context = self._page = None
raise
setattr(self, f"_{self.config.browser_engine}_driver", driver)
return
self._playwright = await async_playwright().start()
self._browser = await self._playwright.chromium.launch(
headless=self.config.headless,
args=[
"--disable-blink-features=AutomationControlled",
"--no-sandbox",
"--disable-setuid-sandbox",
"--disable-infobars",
"--window-size=1280,800",
],
)
self._context = await self._browser.new_context(
viewport={"width": self.config.window_width, "height": self.config.window_height},
user_agent=self.config.user_agent,
device_scale_factor=1,
accept_downloads=True,
locale="en-US",
timezone_id="America/New_York",
permissions=["geolocation", "notifications"],
extra_http_headers={
"Accept-Language": "en-US,en;q=0.9",
"Sec-Ch-Ua": '"Not A(Brand";v="99", "Google Chrome";v="121", "Chromium";v="121"',
"Sec-Ch-Ua-Mobile": "?0",
"Sec-Ch-Ua-Platform": '"Linux"',
},
)
# Stealth CDP Profile: Mask automation signatures & emulate authentic physical browser
await self._context.add_init_script("""
// 1. Mask navigator.webdriver
try {
Object.defineProperty(navigator, 'webdriver', { get: () => undefined });
} catch (e) {}
// 2. Realistic window.chrome object
if (!window.chrome) {
window.chrome = {
app: { isInstalled: false },
runtime: { OnInstalledReason: {}, OnRestartRequiredReason: {} }
};
}
// 3. Realistic plugins list
try {
Object.defineProperty(navigator, 'plugins', {
get: () => [
{ name: 'Chrome PDF Plugin', filename: 'internal-pdf-viewer', description: 'Portable Document Format' },
{ name: 'Chrome PDF Viewer', filename: 'mhjfbmdgcfjbbpaeojofohoefgiehjai', description: '' },
{ name: 'Native Client', filename: 'internal-nacl-plugin', description: '' }
]
});
} catch (e) {}
// 4. Genuine languages
try {
Object.defineProperty(navigator, 'languages', { get: () => ['en-US', 'en'] });
} catch (e) {}
// 5. Hardware-accelerated WebGL vendor profile (NVIDIA)
try {
const origGetParam = WebGLRenderingContext.prototype.getParameter;
WebGLRenderingContext.prototype.getParameter = function(param) {
if (param === 37445) return 'Google Inc. (NVIDIA)';
if (param === 37446) return 'ANGLE (NVIDIA, NVIDIA GeForce RTX 5090 Direct3D11 vs_5_0 ps_5_0, D3D11)';
return origGetParam.apply(this, arguments);
};
} catch (e) {}
""")
self._page = await self._context.new_page()
async def stop(self) -> None:
"""Closes browser context and its driver."""
if self._obscura_driver is not None:
await self._obscura_driver.stop()
self._obscura_driver = None
self._page = self._context = self._browser = None
return
if self._lightpanda_driver is not None:
await self._lightpanda_driver.stop()
self._lightpanda_driver = None
self._page = None
self._context = None
self._browser = None
return
# A stalled page must not prevent closing its browser and transport.
for resource, method in ((self._page, "close"), (self._context, "close"),
(self._browser, "close"), (self._playwright, "stop")):
if resource is not None:
try:
await asyncio.wait_for(getattr(resource, method)(), timeout=5)
except Exception as exc:
logging.getLogger(__name__).warning("Browser cleanup %s failed: %s", method, exc)
self._page = None
self._context = None
self._browser = None
self._playwright = None
@property
def page(self) -> Page:
if self._page is None:
raise RuntimeError("Browser is not started. Call await controller.start() first.")
return self._page
async def goto(self, url: str) -> None:
"""Navigates to URL and waits for DOM content loaded."""
await self.page.goto(url, wait_until="domcontentloaded", timeout=self.config.navigation_timeout_ms)
async def get_pruned_tree(self) -> PrunedAXTree:
"""Extracts the lightweight, sanitized AXTree."""
for attempt in range(3):
try:
return await self.distiller.extract_pruned_tree(self.page)
except PlaywrightError as exc:
if "Execution context was destroyed" not in str(exc) or attempt == 2:
raise
await self.page.wait_for_load_state("domcontentloaded", timeout=self.config.navigation_timeout_ms)
await asyncio.sleep(0.2)
async def screenshot(self, output_path: Union[Path, str]) -> Path:
"""Captures viewport screenshot for vision grounding or audit trail."""
output_path = Path(output_path)
output_path.parent.mkdir(parents=True, exist_ok=True)
await self.page.screenshot(path=str(output_path), full_page=False)
return output_path
async def capture_annotated_step_screenshot(
self,
output_path: Union[Path, str],
step_num: int,
action_label: str,
bbox: Optional[Tuple[float, float, float, float]] = None,
click_coords: Optional[Tuple[float, float]] = None,
) -> Path:
"""
Captures a high-resolution screenshot with visual annotations:
- Target element bounding box with translucent green tint
- Click coordinate glowing crosshairs
- Step execution HUD badge
"""
output_path = Path(output_path)
output_path.parent.mkdir(parents=True, exist_ok=True)
raw_bytes = await self.page.screenshot(full_page=False)
img = Image.open(io.BytesIO(raw_bytes)).convert("RGBA")
overlay = Image.new("RGBA", img.size, (0, 0, 0, 0))
draw = ImageDraw.Draw(overlay)
# 1. Bounding box overlay
if bbox and bbox[2] > 0 and bbox[3] > 0:
bx, by, bw, bh = bbox
draw.rectangle(
[bx, by, bx + bw, by + bh],
fill=(16, 185, 129, 45),
outline=(16, 185, 129, 255),
width=3,
)
# 2. Click coordinate crosshair overlay
if click_coords:
cx, cy = click_coords
# Outer ring
draw.ellipse([cx - 14, cy - 14, cx + 14, cy + 14], outline=(239, 68, 68, 220), width=3)
# Center target
draw.ellipse([cx - 4, cy - 4, cx + 4, cy + 4], fill=(239, 68, 68, 255))
# Crosshair tick marks
draw.line([cx - 20, cy, cx - 6, cy], fill=(239, 68, 68, 240), width=2)
draw.line([cx + 6, cy, cx + 20, cy], fill=(239, 68, 68, 240), width=2)
draw.line([cx, cy - 20, cx, cy - 6], fill=(239, 68, 68, 240), width=2)
draw.line([cx, cy + 6, cx, cy + 20], fill=(239, 68, 68, 240), width=2)
# 3. Top-left HUD badge
badge_width = min(420, max(280, len(action_label) * 8 + 60))
draw.rectangle([12, 12, 12 + badge_width, 50], fill=(15, 23, 42, 225), outline=(56, 189, 248, 255), width=2)
draw.text((22, 20), f"Step {step_num}: {action_label[:45]}", fill=(248, 250, 252, 255))
annotated = Image.alpha_composite(img, overlay).convert("RGB")
annotated.save(output_path, "PNG")
return output_path
async def execute_action(self, action: BrowserAction, tree: Optional[PrunedAXTree] = None) -> Dict[str, Any]:
"""Executes a single atomic action."""
return await self.executor.execute_action(self.page, action, tree)
async def execute_batch(self, batch: ActionBatch, tree: Optional[PrunedAXTree] = None) -> List[Dict[str, Any]]:
"""Executes a speculative batch of actions."""
return await self.executor.execute_batch(self.page, batch, tree)
async def execute_chunk(
self,
actions: List[BrowserAction],
tree: Optional[PrunedAXTree] = None,
early_stop_on_navigation: bool = True,
) -> List[Dict[str, Any]]:
"""Executes a pipelined speculative action chunk."""
return await self.executor.execute_chunk(
self.page, actions, tree, early_stop_on_navigation=early_stop_on_navigation
)
async def scrape_page(self) -> Dict[str, Any]:
"""Scrapes structured page data (metadata, tables, card elements, text)."""
action = BrowserAction(op=BrowserActionType.EXTRACT)
details = await self.executor.execute_action(self.page, action)
return details.get("extracted_data", {})
async def scrape_element(self, target: str, tree: Optional[PrunedAXTree] = None) -> Dict[str, Any]:
"""Scrapes structured content and attributes from a specific targeted element."""
action = BrowserAction(op=BrowserActionType.EXTRACT, target=target)
details = await self.executor.execute_action(self.page, action, tree)
return details.get("extracted_data", {})
async def scrape_table(self, selector: str = "table") -> List[List[str]]:
"""Scrapes 2D matrix rows from a tabular element."""
action = BrowserAction(op=BrowserActionType.EXTRACT, extract_selector=selector)
details = await self.executor.execute_action(self.page, action)
data = details.get("extracted_data", [])
return data
|