Upload 2 files

app.py

@@ -7,12 +7,12 @@ from pathlib import Path
 from huggingface_hub import hf_hub_download, list_repo_files
 
 
-MODEL_REPO = os.getenv("MODEL_REPO", "ggml-org/SmolVLM2-2.2B-Instruct-GGUF")
+MODEL_REPO = os.getenv("MODEL_REPO", "mradermacher/Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning-i1-GGUF")
 QUANT = os.getenv("QUANT", "Q4_K_M")            # used to auto-pick the model file
 MODEL_FILE = os.getenv("MODEL_FILE", "auto")     # or an exact filename
-MMPROJ_FILE = os.getenv("MMPROJ_FILE", "auto")   # or an exact filename, or "none" for text-only
+MMPROJ_FILE = os.getenv("MMPROJ_FILE", "none")   # text-only model: no projector needed
 MMPROJ_PREFER = os.getenv("MMPROJ_PREFER", "Q8_0,f16,F16")  # auto-pick order
-MODEL_DIR = Path(os.getenv("MODEL_DIR", "/data/models/smolvlm2"))
+MODEL_DIR = Path(os.getenv("MODEL_DIR", "/data/models/llama33-8b-thinking"))
 
 LLAMA_SERVER_BIN = os.getenv("LLAMA_SERVER_BIN", "/opt/llama.cpp/llama-server")
 LLAMA_HOST = os.getenv("LLAMA_HOST", "0.0.0.0")
@@ -29,10 +29,10 @@ CACHE_TYPE_V = os.getenv("CACHE_TYPE_V", "default")
 IMAGE_MAX_TOKENS = os.getenv("IMAGE_MAX_TOKENS", "default")  # only for dynamic-resolution models
 CHAT_TEMPLATE_FILE = os.getenv("CHAT_TEMPLATE_FILE", "default")  # default = use template embedded in GGUF
 
-TEMPERATURE = os.getenv("TEMPERATURE", "0.2")
+TEMPERATURE = os.getenv("TEMPERATURE", "0.6")
 TOP_P = os.getenv("TOP_P", "0.95")
-TOP_K = os.getenv("TOP_K", "64")
-REPEAT_PENALTY = os.getenv("REPEAT_PENALTY", "1.08")  # keep 1.0 for OCR / vision
+TOP_K = os.getenv("TOP_K", "40")
+REPEAT_PENALTY = os.getenv("REPEAT_PENALTY", "1.05")
 
 
 def log(message: str) -> None: