Spaces:
Sleeping
Sleeping
Update app_deepseek.py
Browse files- app_deepseek.py +2 -2
app_deepseek.py
CHANGED
|
@@ -388,8 +388,8 @@ SEARCH_LANGUAGE_CYCLE = ["arabic", "english"]
|
|
| 388 |
# "llamacpp" (default): llama-server + GGUF, on GPU (CUDA build) or CPU. llama.cpp cannot use TPU.
|
| 389 |
# "hf": the old transformers path (only kept for reference; on TPU it produced ~3 chars/s and hung).
|
| 390 |
BACKEND = os.environ.get("GIJL_BACKEND", "llamacpp").strip().lower()
|
| 391 |
-
GGUF_REPO = os.environ.get("GIJL_GGUF_REPO", "unsloth/DeepSeek-V4-Flash-
|
| 392 |
-
GGUF_QUANT = os.environ.get("GIJL_GGUF_QUANT", "UD-
|
| 393 |
# automatically to whatever fits the detected VRAM/RAM
|
| 394 |
# budget (see _memory_budget_gb) -- set GIJL_GGUF_QUANT
|
| 395 |
# explicitly to pin a smaller quant instead.
|
|
|
|
| 388 |
# "llamacpp" (default): llama-server + GGUF, on GPU (CUDA build) or CPU. llama.cpp cannot use TPU.
|
| 389 |
# "hf": the old transformers path (only kept for reference; on TPU it produced ~3 chars/s and hung).
|
| 390 |
BACKEND = os.environ.get("GIJL_BACKEND", "llamacpp").strip().lower()
|
| 391 |
+
GGUF_REPO = os.environ.get("GIJL_GGUF_REPO", "unsloth/DeepSeek-V4-Flash-GGUF")
|
| 392 |
+
GGUF_QUANT = os.environ.get("GIJL_GGUF_QUANT", "UD-Q2_K_XL") # highest quant unsloth ships; choose_quant() steps this down
|
| 393 |
# automatically to whatever fits the detected VRAM/RAM
|
| 394 |
# budget (see _memory_budget_gb) -- set GIJL_GGUF_QUANT
|
| 395 |
# explicitly to pin a smaller quant instead.
|