docs: vicuna example

docsite/docs/fal-serverless/examples/llm_vicuna_example.md

@@ -0,0 +1,60 @@
+```
+from fal_serverless import isolated, cached
+
+requirements = [
+"transformers",
+"sentencepiece",
+"accelerate",
+"fal_serverless",
+"torch==2.0",
+"numpy",
+"tokenizers>=0.12.1",
+]
+
+@cached
+def load_model():
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+    print(torch.__version__)
+
+    tokenizer = AutoTokenizer.from_pretrained(
+        "TheBloke/vicuna-13B-1.1-HF", use_fast=False
+    )
+
+    model = AutoModelForCausalLM.from_pretrained(
+        "TheBloke/vicuna-13B-1.1-HF", low_cpu_mem_usage=True, torch_dtype=torch.float16
+    )
+
+    model.to("cuda")
+
+    return tokenizer, model
+
+@isolated(requirements=requirements, machine_type="GPU", keep_alive=60)
+def llm(prompt):
+import os
+import torch
+
+    print(torch.__version__)
+
+    os.environ["TRANSFORMERS_CACHE"] = "/data/hfcache"
+
+    with torch.inference_mode():
+        tokenizer, model = load_model()
+        input_ids = tokenizer([prompt]).input_ids
+
+        output_ids = model.generate(
+            torch.as_tensor(input_ids).cuda(),
+            do_sample=True,
+            temperature=0.7,
+            max_new_tokens=512,
+        )
+        output_ids = output_ids[0][len(input_ids[0]) :]
+        outputs = tokenizer.decode(
+            output_ids, skip_special_tokens=True, spaces_between_special_tokens=False
+        )
+
+        return outputs
+
+print(llm("give me a itenary for a trip to japan"))
+```