Update README.md

README.md

@@ -10,10 +10,9 @@ license: apache-2.0
 
 slim-sentiment has been fine-tuned for **sentiment analysis** function calls, with output of JSON dictionary corresponding to specific named entity keys.  
 
-Each slim model has a corresponding 'tool' in a separate repository, e.g., 'slim-sentiment-tool', which a 4-bit quantized gguf version of the model that is intended to be used for inference.  
-
-
+Each slim model has a corresponding 'tool' in a separate repository, e.g., [**'slim-sentiment-tool'**](www.huggingface.co/llmware/slim-sentiment-tool/), which a 4-bit quantized gguf version of the model that is intended to be used for inference.  
 
+Inference speed and loading time is much faster with the 'tool' versions of the model.   
 
 ### Model Description
 
@@ -48,31 +47,61 @@ All of the SLIM models use a novel prompt instruction structured as follows:
 
 The fastest way to get started with BLING is through direct import in transformers:
 
-from transformers import AutoTokenizer, AutoModelForCausalLM  
-tokenizer = AutoTokenizer.from_pretrained("slim-sentiment")  
-model = AutoModelForCausalLM.from_pretrained("slim-sentiment")  
+    import ast
+    from transformers import AutoModelForCausalLM, AutoTokenizer
+    
+    model = AutoModelForCausalLM.from_pretrained("llmware/slim-sentiment")
+    tokenizer = AutoTokenizer.from_pretrained("llmware/slim-sentiment")
+
+    text = "The markets declined for a second straight days on news of disappointing earnings."
+
+    keys = "sentiment"
+
+    prompt = "<human>: " + text + "\n" + "<classify> " + keys + "</classify>" + "\n<bot>: "
+
+    #   huggingface standard generation script
+    inputs = tokenizer(prompt, return_tensors="pt")
+    start_of_output = len(inputs.input_ids[0])
+
+    outputs = model.generate(inputs.input_ids.to('cpu'), eos_token_id=tokenizer.eos_token_id,
+                             pad_token_id=tokenizer.eos_token_id, do_sample=True, temperature=0.3, max_new_tokens=100)
 
+    output_only = tokenizer.decode(outputs[0][start_of_output:], skip_special_tokens=True)
 
-The BLING model was fine-tuned with a simple "\<human> and \<bot> wrapper", so to get the best results, wrap inference entries as:
+    print("input text sample - ", text)
+    print("llm_response - ", output_only)
 
-full_prompt = "\<human>\: " + my_prompt + "\n" + "\<bot>\:"
+    #   where it gets interesting
+    try:
+        # convert llm response output from string to json
+        output_only = ast.literal_eval(output_only)
+        print("converted to json automatically")
 
-The BLING model was fine-tuned with closed-context samples, which assume generally that the prompt consists of two sub-parts:
+        #   look for the key passed in the prompt as a dictionary entry
+        if keys in output_only:
+            if "negative" in output_only[keys]:
+                print("sentiment appears negative - need to handle ...")
+        else:
+            print("response does not appear to include the designated key - will need to try again.")
 
-1.  Text Passage Context, and
-2.  Specific question or instruction based on the text passage
+    except:
+        print("could not convert to json automatically - ", output_only)
 
-To get the best results, package "my_prompt" as follows:
 
-my_prompt = {{text_passage}} + "\n" + {{question/instruction}}
+## Using as Function Call in LLMWare
 
+We envision the slim models deployed in a pipeline/workflow/templating framework that handles the prompt packaging more elegantly.  Check out llmware for one such implementation:  
 
+    from llmware.models import ModelCatalog
+    slim_model = ModelCatalog().load_model("llmware/slim-sentiment")
+    response = slim_model.function_call(text,params=["sentiment"], function="classify")
 
+    print("llmware - llm_response: ", response)
+
+    
 ## Model Card Contact
 
 Darren Oberst & llmware team
 
-Please reach out anytime if you are interested in this project and would like to participate and work with us!
-