Spaces:
Paused
Paused
File size: 5,636 Bytes
d072c3b | 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 | """Modelifier. Point at a model, add water, test the cut, create the model."""
import json
import os
import gradio as gr
from modelifier.surgeon import inspect_model, publish, test_edit
INK = "#0b0b0c"
PAPER = "#e7e2d6"
SIGNAL = "#7dd3c0"
MUTED = "#8b9098"
LINE = "#2c2c31"
CSS = f"""
body, .gradio-container {{ background: {INK}; color: {PAPER}; }}
.gradio-container {{ max-width: 980px; margin: 0 auto; }}
footer {{ display: none !important; }}
"""
THEME = gr.themes.Base(
primary_hue=gr.themes.colors.teal,
secondary_hue=gr.themes.colors.stone,
neutral_hue=gr.themes.colors.stone,
font=gr.themes.GoogleFont("IBM Plex Sans"),
font_mono=gr.themes.GoogleFont("IBM Plex Mono"),
).set(body_background_fill=INK, body_text_color=PAPER, block_background_fill="#141416")
def _layers(value: str):
value = (value or "").strip()
if not value:
return None
return [int(part.strip()) for part in value.split(",") if part.strip()]
def _water(value: str):
return [line.strip() for line in (value or "").splitlines() if line.strip()]
def run_inspect(pointer: str, token: str):
try:
report = inspect_model(pointer, token.strip() or None)
except Exception as exc:
raise gr.Error(str(exc)) from exc
lines = [
f"{report['repo_id']}",
f"files {report['files']} · hidden {report['hidden_size']} · layers {report['layers']}",
f"embedding {report['embedding']}",
"weights not downloaded",
"",
]
for row in report["resolved"]:
lines.append(f"L{row['layer']} {row['tensor']} {row['shard']}")
return "\n".join(lines), json.dumps(report, indent=2)
def run_test(pointer, water, note, layers, alpha, top_k, token):
try:
result = test_edit(
pointer,
_water(water),
note or "",
_layers(layers),
float(alpha),
int(top_k),
token.strip() or None,
)
except Exception as exc:
raise gr.Error(str(exc)) from exc
lines = [
f"{result['source']} · not published",
f"water samples {result['water_count']} · alpha {result['alpha']} · expected scale {1 - result['alpha']:.2f}",
"",
]
for row in result["layers"]:
lines.append(
f"L{row['layer']} {row['tensor']} cols {row['columns']} "
f"scale {row['component_scale']} ΔF {row['relative_frobenius_pct']}% "
f"σ2/σ1 {row['sigma2_over_sigma1_pct']}%"
)
lines.append("")
lines.append("sha changed on every touched tensor. nothing was created.")
return "\n".join(lines), json.dumps(result, indent=2)
def run_publish(pointer, water, note, layers, alpha, top_k, name, chat, private, token):
if not token.strip():
raise gr.Error("A write token is required to create the model. Test does not need one.")
if not name.strip():
raise gr.Error("Name the model.")
try:
result = publish(
pointer,
_water(water),
name.strip(),
note or "",
_layers(layers),
float(alpha),
int(top_k),
bool(chat),
bool(private),
token.strip(),
)
except Exception as exc:
raise gr.Error(str(exc)) from exc
lines = [
result["model_url"],
result.get("space_url") or "chat Space not requested",
f"account {result['account']}",
]
return "\n".join(lines), json.dumps({k: result[k] for k in ("model", "model_url", "space_url", "layers")}, indent=2)
with gr.Blocks(theme=THEME, css=CSS, title="Modelifier") as demo:
gr.Markdown(
"""
# Modelifier
Point at a model. Point at datasets or other models as water. The cut is
`wⱼ ← wⱼ − α (uᵀ wⱼ) u` on the residual out-projection. Test writes nothing.
Create needs a write token and makes the named model, plus a chat Space if you ask.
"""
)
pointer = gr.Textbox(label="Model", placeholder="https://huggingface.co/org/model")
water = gr.Textbox(
label="Water — one model or dataset URL per line",
lines=4,
placeholder="https://huggingface.co/datasets/org/set",
)
note = gr.Textbox(label="What you want in the water", lines=2)
with gr.Row():
layers = gr.Textbox(label="Layers (blank = middle)", placeholder="15, 19")
alpha = gr.Slider(0, 12, value=1, step=0.5, label="α · 1 removes, 6 flips")
top_k = gr.Slider(4, 256, value=64, step=4, label="Columns K")
token = gr.Textbox(label="Write token — create only", type="password")
with gr.Row():
inspect_btn = gr.Button("Read the model")
test_btn = gr.Button("Test the cut", variant="primary")
out = gr.Textbox(label="Result", lines=10)
detail = gr.Code(label="Audit", language="json")
inspect_btn.click(run_inspect, [pointer, token], [out, detail])
test_btn.click(run_test, [pointer, water, note, layers, alpha, top_k, token], [out, detail])
gr.Markdown("### Create under your account")
with gr.Row():
name = gr.Textbox(label="Model name", placeholder="qwen-terse")
chat = gr.Checkbox(value=True, label="Also create a chat Space")
private = gr.Checkbox(value=False, label="Private")
create_btn = gr.Button("Create model", variant="primary")
created = gr.Textbox(label="Created", lines=4)
create_btn.click(
run_publish,
[pointer, water, note, layers, alpha, top_k, name, chat, private, token],
[created, detail],
)
if __name__ == "__main__":
demo.launch()
|