Safetensors
GGUF
English
Russian
lfm3
reasoning
test-time-compute
inference-time-scaling
code-generation
llama.cpp
negative-results
conversational
Instructions to use HelioAI/LFM3-1.7B-inference-modules-experiments with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M # Run inference directly in the terminal: llama cli -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M # Run inference directly in the terminal: llama cli -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Use Docker
docker model run hf.co/HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
- Llama new
- LM Studio
- Jan
- Ollama
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with Ollama:
ollama run hf.co/HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
- Unsloth Desktop
- Pi
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with Docker Model Runner:
docker model run hf.co/HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
- Lemonade
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Run and chat with the model
lemonade run user.LFM3-1.7B-inference-modules-experiments-Q5_K_M
List all available models
lemonade list
- Hermes Agent
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use HelioAI/LFM3-1.7B-inference-modules-experiments with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "HelioAI/LFM3-1.7B-inference-modules-experiments:Q5_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
File size: 8,461 Bytes
67ebcef | 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 | """Text 'eyes' without any model: what a rendered page shows, read from the browser's layout (DOM) and, for
drawings, from the geometry of the SVG elements. Used as feedback for the model, which cannot see its result.
outline(page) -> lines like "first screen: navbar 'Golden Crust | Home About Shop'", "y 820-1100 (below the first
screen): form with inputs Email, Password; checkbox 'Remember me'; button 'Sign in'".
svg_outline(page) -> the largest shapes and all texts with their positions and colours, back to front.
"""
OUTLINE_JS = r"""
() => {
const H = innerHeight, out = [], seen = new Set();
const vis = (el) => { const r = el.getBoundingClientRect(), s = getComputedStyle(el);
return r.width > 3 && r.height > 3 && s.visibility !== 'hidden' && s.display !== 'none' && +s.opacity > 0.05; };
const txt = (el, n = 60) => (el.innerText || el.value || el.placeholder || el.getAttribute('aria-label') || '')
.replace(/\s+/g, ' ').trim().slice(0, n);
const where = (r) => (r.top < H - 40 ? 'first screen' : `y ${Math.round(r.top + scrollY)} (below the first screen)`);
const colour = (el) => getComputedStyle(el).backgroundColor;
const sel = 'nav,header,footer,aside,h1,h2,h3,form,table,button,a[class*=btn],input,select,textarea,img,canvas,svg,' +
'[class*=card],[class*=modal],[class*=hero],[class*=sidebar],details,blockquote,pre,ul,ol';
for (const el of document.querySelectorAll(sel)) {
if (!vis(el)) continue;
if ([...seen].some((p) => p.contains(el) && p.matches('form,table,ul,ol,nav,[class*=card],details,pre,blockquote'))) continue;
const r = el.getBoundingClientRect(), tag = el.tagName.toLowerCase(), cls = (el.className && el.className.baseVal === undefined) ? String(el.className) : '';
let kind = tag, detail = txt(el);
if (tag === 'form') {
const parts = [...el.querySelectorAll('input,select,textarea,button')].filter(vis).map((x) => {
const t = x.type || x.tagName.toLowerCase(), lab = x.labels && x.labels[0] ? x.labels[0].innerText.trim() : (x.placeholder || x.name || x.value || x.innerText || '').trim();
return `${t === 'submit' || x.tagName === 'BUTTON' ? 'button' : t} '${lab.slice(0, 30)}'`; });
detail = parts.join('; ');
} else if (tag === 'table') {
const rows = el.rows.length, head = [...(el.rows[0] || { cells: [] }).cells].map((c) => c.innerText.trim()).join(' | ');
detail = `${rows} rows, header: ${head.slice(0, 80)}`;
} else if (tag === 'ul' || tag === 'ol') {
detail = `${el.children.length} items: ` + [...el.children].slice(0, 4).map((li) => li.innerText.trim().slice(0, 25)).join(' / ');
} else if (/card/.test(cls)) { kind = 'card'; }
else if (/modal/.test(cls)) { kind = 'modal (open, covers the page)'; }
else if (/hero/.test(cls)) { kind = 'hero section'; detail = txt(el, 80); }
else if (tag === 'img' || tag === 'canvas' || tag === 'svg') { detail = `${Math.round(r.width)}x${Math.round(r.height)} px`; }
else if (tag === 'input') { detail = `${el.type} '${txt(el, 30)}'`; }
seen.add(el);
out.push(`${where(r)}: ${kind}${detail ? " '" + detail + "'" : ''} [${Math.round(r.left)},${Math.round(r.top + scrollY)} ${Math.round(r.width)}x${Math.round(r.height)}, bg ${colour(el)}]`);
if (out.length >= 40) break;
}
const h = document.documentElement.scrollHeight;
out.unshift(`page height ${h} px (the first screen is 1024x768); body background ${getComputedStyle(document.body).backgroundColor}`);
return out;
}
"""
SVG_JS = r"""
() => {
const svg = document.querySelector('svg'); if (!svg) return ['no <svg> element'];
const vb = svg.viewBox && svg.viewBox.baseVal && svg.viewBox.baseVal.width ? svg.viewBox.baseVal : { x: 0, y: 0, width: svg.clientWidth, height: svg.clientHeight };
const out = [`viewBox ${vb.width}x${vb.height}`], items = [];
svg.querySelectorAll('rect,circle,ellipse,path,polygon,polyline,line,text').forEach((el, i) => {
let b; try { b = el.getBBox(); } catch (e) { return; }
if (!b || (b.width < 0.5 && b.height < 0.5)) return;
const s = getComputedStyle(el), fill = s.fill, stroke = s.stroke;
const area = b.width * b.height, outside = b.x + b.width < vb.x || b.y + b.height < vb.y || b.x > vb.x + vb.width || b.y > vb.y + vb.height;
items.push({ i, tag: el.tagName, x: Math.round(b.x), y: Math.round(b.y), w: Math.round(b.width), h: Math.round(b.height),
fill, stroke, text: el.tagName === 'text' ? el.textContent.trim().slice(0, 30) : '', area, outside });
});
const big = items.filter((o) => o.tag !== 'text').sort((a, b) => b.area - a.area).slice(0, 18).sort((a, b) => a.i - b.i);
for (const o of big) out.push(`${o.tag} at x ${o.x}-${o.x + o.w}, y ${o.y}-${o.y + o.h}, fill ${o.fill}${o.stroke !== 'none' ? ', stroke ' + o.stroke : ''}${o.outside ? ' (OUTSIDE the picture)' : ''}`);
for (const o of items.filter((o) => o.tag === 'text')) out.push(`text '${o.text}' at x ${o.x}-${o.x + o.w}, y ${o.y}-${o.y + o.h}`);
out.push(`${items.length} shapes in total (back to front in the order listed; y grows downwards)`);
return out;
}
"""
def outline(page):
return page.evaluate(OUTLINE_JS)
def svg_outline(page):
return page.evaluate(SVG_JS)
NAMES = {"black": (20, 20, 20), "dark grey": (70, 70, 70), "grey": (140, 140, 140), "light grey": (200, 200, 200),
"white": (245, 245, 245), "red": (220, 40, 40), "dark red": (130, 20, 20), "orange": (240, 140, 30),
"yellow": (240, 220, 40), "brown": (130, 80, 40), "green": (50, 170, 60), "dark green": (20, 90, 30),
"light green": (150, 220, 120), "teal": (30, 160, 160), "light blue": (140, 200, 240), "blue": (40, 90, 220),
"dark blue": (20, 30, 90), "purple": (130, 60, 190), "pink": (240, 120, 180), "beige": (235, 220, 190)}
def colour_name(rgb):
return min(NAMES, key=lambda k: sum((a - b) ** 2 for a, b in zip(NAMES[k], rgb)))
def pixel_outline(png_bytes, png2_bytes=None):
"""For canvas / WebGL pages: the background colour and the coloured regions of the screenshot (connected blobs on
a 64x48 grid), with position, size and colour; and which regions moved between two screenshots."""
import io
import numpy as np
from PIL import Image
def grid(b):
return np.asarray(Image.open(io.BytesIO(b)).convert("RGB").resize((64, 48), Image.BILINEAR)).astype(int)
a = grid(png_bytes)
flat = a.reshape(-1, 3)
vals, counts = np.unique((flat // 24) * 24, axis=0, return_counts=True)
bg = vals[counts.argmax()] + 12
fg = np.abs(a - bg).sum(axis=2) > 60
out = [f"background {colour_name(bg)}; {100 * fg.mean():.0f}% of the screen is covered by drawn content"]
seen, blobs = np.zeros_like(fg), []
for y in range(48):
for x in range(64):
if fg[y, x] and not seen[y, x]:
stack, cells = [(y, x)], []
seen[y, x] = True
while stack:
cy, cx = stack.pop()
cells.append((cy, cx))
for ny, nx in ((cy + 1, cx), (cy - 1, cx), (cy, cx + 1), (cy, cx - 1)):
if 0 <= ny < 48 and 0 <= nx < 64 and fg[ny, nx] and not seen[ny, nx]:
seen[ny, nx] = True
stack.append((ny, nx))
blobs.append(cells)
blobs.sort(key=len, reverse=True)
def pos(cy, cx):
v = "top" if cy < 16 else "bottom" if cy >= 32 else "middle"
h = "left" if cx < 21 else "right" if cx >= 43 else "centre"
return "centre" if (v, h) == ("middle", "centre") else f"{v} {h}" if v != "middle" else f"middle {h}"
for cells in blobs[:10]:
if len(cells) < 4:
break
ys, xs = [c[0] for c in cells], [c[1] for c in cells]
col = np.mean([a[c] for c in cells], axis=0)
out.append(f"{colour_name(col)} shape at the {pos(np.mean(ys), np.mean(xs))}, about {100 * (max(xs) - min(xs) + 1) / 64:.0f}% "
f"of the width x {100 * (max(ys) - min(ys) + 1) / 48:.0f}% of the height ({100 * len(cells) / 3072:.1f}% of the screen)")
small = sum(1 for c in blobs if len(c) < 4)
if small:
out.append(f"{small} small dots or specks")
if png2_bytes is not None:
moved = (np.abs(grid(png2_bytes) - a).sum(axis=2) > 60).mean()
out.append(f"{100 * moved:.1f}% of the screen changed in 0.7 seconds" + (" (nothing moves)" if moved < 0.002 else ""))
return out
|