Spaces:
Running
Running
Q5
Browse files
app.py
CHANGED
|
@@ -27,6 +27,7 @@ QUANT_PARAMS = {
|
|
| 27 |
"Q2": 2,
|
| 28 |
"Q3": 3,
|
| 29 |
"Q4": 4,
|
|
|
|
| 30 |
"Q6": 6,
|
| 31 |
"Q8": 8,
|
| 32 |
}
|
|
@@ -144,6 +145,7 @@ def process_model(model_id, q_method, oauth_token: gr.OAuthToken | None):
|
|
| 144 |
css="""/* Custom CSS to allow scrolling */
|
| 145 |
.gradio-container {overflow-y: auto;}
|
| 146 |
"""
|
|
|
|
| 147 |
# Create Gradio interface
|
| 148 |
with gr.Blocks(css=css) as demo:
|
| 149 |
gr.Markdown("You must be logged in to use MLX-my-repo.")
|
|
@@ -156,7 +158,7 @@ with gr.Blocks(css=css) as demo:
|
|
| 156 |
)
|
| 157 |
|
| 158 |
q_method = gr.Dropdown(
|
| 159 |
-
["FP16"
|
| 160 |
label="Conversion Method",
|
| 161 |
info="MLX conversion type (FP16 for float16, Q2–Q8 for quantized models)",
|
| 162 |
value="Q4",
|
|
|
|
| 27 |
"Q2": 2,
|
| 28 |
"Q3": 3,
|
| 29 |
"Q4": 4,
|
| 30 |
+
"Q5": 5,
|
| 31 |
"Q6": 6,
|
| 32 |
"Q8": 8,
|
| 33 |
}
|
|
|
|
| 145 |
css="""/* Custom CSS to allow scrolling */
|
| 146 |
.gradio-container {overflow-y: auto;}
|
| 147 |
"""
|
| 148 |
+
|
| 149 |
# Create Gradio interface
|
| 150 |
with gr.Blocks(css=css) as demo:
|
| 151 |
gr.Markdown("You must be logged in to use MLX-my-repo.")
|
|
|
|
| 158 |
)
|
| 159 |
|
| 160 |
q_method = gr.Dropdown(
|
| 161 |
+
["FP16"] + list(QUANT_PARAMS.keys()),
|
| 162 |
label="Conversion Method",
|
| 163 |
info="MLX conversion type (FP16 for float16, Q2–Q8 for quantized models)",
|
| 164 |
value="Q4",
|