Spaces:
Running
Running
Commit ·
83dda90
1
Parent(s): 540645f
fix(app): resolve powershell string syntax error
Browse files
app.py
CHANGED
|
@@ -38,11 +38,13 @@ try:
|
|
| 38 |
LOCAL_7B_P95 = 37.47
|
| 39 |
|
| 40 |
if LOCAL_7B_P95 <= 14.0:
|
| 41 |
-
local_repo = "bartowski/Qwen2.5-7B-Instruct-GGUF"
|
|
|
|
| 42 |
LOCAL_MAX_TOKENS = 150
|
| 43 |
logger.info(f"Using 7b for local tier since p95 latency ({LOCAL_7B_P95}s) is within 14s budget.")
|
| 44 |
else:
|
| 45 |
-
local_repo = "Qwen/Qwen2.5-3B-Instruct-GGUF"
|
|
|
|
| 46 |
LOCAL_MAX_TOKENS = 250
|
| 47 |
logger.info(f"7b p95 {LOCAL_7B_P95}s exceeds 14s budget, using 3b instead.")
|
| 48 |
|
|
|
|
| 38 |
LOCAL_7B_P95 = 37.47
|
| 39 |
|
| 40 |
if LOCAL_7B_P95 <= 14.0:
|
| 41 |
+
local_repo = "bartowski/Qwen2.5-7B-Instruct-GGUF"
|
| 42 |
+
local_file = "Qwen2.5-7B-Instruct-Q4_K_M.gguf"
|
| 43 |
LOCAL_MAX_TOKENS = 150
|
| 44 |
logger.info(f"Using 7b for local tier since p95 latency ({LOCAL_7B_P95}s) is within 14s budget.")
|
| 45 |
else:
|
| 46 |
+
local_repo = "Qwen/Qwen2.5-3B-Instruct-GGUF"
|
| 47 |
+
local_file = "qwen2.5-3b-instruct-q4_k_m.gguf"
|
| 48 |
LOCAL_MAX_TOKENS = 250
|
| 49 |
logger.info(f"7b p95 {LOCAL_7B_P95}s exceeds 14s budget, using 3b instead.")
|
| 50 |
|