fix: stbalelm

requirements.txt

@@ -2,6 +2,11 @@
 # chmod +x flatpak-pip-generator
 # ./flatpak-pip-generator --requirements-file=requirements.txt --output pypi-dependencies
 
+# use https://johannesjh.github.io/req2flatpak/main/cli.html
+
+# 1. pip-compile -o r.txt requirements.txt
+# 2. ./req2flatpak.py --requirements-file requirements.txt --target-platforms  310-x86_64 310-aarch64 > build-aux/pypi-dependencies.json
+
 Pygments
 baichat_py
 googlebardpy

src/provider/__init__.py

@@ -15,7 +15,7 @@ from .hfdialogpt import HuggingFaceDialoGPTLargeProvider
 from .hfgpt2large import HuggingFaceGPT2LargeProvider
 from .hfgpt2xl import HuggingFaceGPT2XLProvider
 from .stablelm import StableLMProvider
-# from .starcoder import StarCoderProvider
+from .starcoder import StarCoderProvider
 
 PROVIDERS = {
     "alpacalora": AlpacaLoRAProvider,

src/provider/gradio.py

@@ -19,12 +19,13 @@ class BaseGradioProvider(BavarderProvider):
         self.client = Client(self.url)
 
     def ask(self, prompt):
+        print("ASKING " + "="*100)
         try:
-            result = client.predict(
+            response = self.client.predict(
 				prompt,	# str representing string value in 'Chat Message Box' Textbox component
-				"null",	# str representing filepath to JSON file in 'parameter_3' Chatbot component
 				fn_index=0
             )
+            print(response)
         except Exception as e:
             print(e)
             self.win.banner.props.title = str(e)

src/provider/stablelm.py

@@ -1,6 +1,6 @@
-from .gradio import BaseGradioProvider
+from .huggingface import BaseHFProvider
 
-class StableLMProvider(BaseGradioProvider):
+class StableLMProvider(BaseHFProvider):
     name = "StableLM"
     slug = "stablelm"
-    url = "https://stabilityai-stablelm-tuned-alpha-chat.hf.space/"
+    model = "stabilityai/stablelm-tuned-alpha-3b"

src/provider/transformer.py

@@ -13,14 +13,18 @@ class BaseTransformerProvider(BavarderProvider):
     slug = None
     checkpoint = None
     device = "cpu"
+    is_setup = False
 
     def __init__(self, win, app, *args, **kwargs):
         super().__init__(win, app, *args, **kwargs)
 
+    def setup(self):
         self.tokenizer = AutoTokenizer.from_pretrained(self.checkpoint)
         self.model = AutoModelForCausalLM.from_pretrained(self.checkpoint).to(self.device)
 
     def ask(self, prompt):
+        if not self.is_setup:
+            self.setup()
         try:
             inputs = tokenizer.encode(prompt, return_tensors="pt").to(self.device)
             outputs = model.generate(inputs)