update

app.py

@@ -2,10 +2,12 @@ import os
 # os.environ["HF_HOME"] = "/data/.huggingface"
 import gradio as gr
 import sys
+import copy
 from datasets import load_dataset
 from typing import List
-from tigerscore import TIGERScorer
-
+from llama_cpp import Llama
+from huggingface_hub import hf_hub_download  
+from string import Template
 
 DESCRIPTIONS = """
 We present ***TIGERScore***, a **T**rained metric that follows **I**nstruction **G**uidance to perform **E**xplainable, and **R**eference-free evaluation over a wide spectrum of text generation tasks. Different from other automatic evaluation methods that only provide arcane scores, TIGERScore is guided by the natural language instruction to provide error analysis to pinpoint the mistakes in the generated text.
@@ -24,18 +26,56 @@ for i, ex in enumerate(SHUFFLED_EXAMPLES_DATASET.take(100)):
     #     continue
     EXAMPLES.append([ex[field] for field in fields])
 
-scorer = TIGERScorer("TIGER-Lab/TIGERScore-13B-GGUF", use_llamacpp=True)
-
-def submit_fn(input_context, generation_instruction, hypo_output, max_new_tokens=512, temperature=0.7, top_p=1.0):
-    return scorer.score(
-        insts=[generation_instruction],
-        hypo_outputs=[hypo_output],
-        input_contexts=[input_context],
-        max_new_tokens=max_new_tokens,
-        temperature=temperature,
-        top_p=top_p,
-    )[0]['raw_output'].strip() 
-        
+TEMPLATE = """You are evaluating errors in a model-generated output for a given instruction.
+Instruction: 
+${generation_instruction}
+${input_context}
+
+Model-generated Output: 
+${hypothesis_output}
+
+For each error you give in the response, please also elaborate the following information:
+- error location (the words that are wrong in the output)
+- error aspect it belongs to.
+- explanation why it's an error, and the correction suggestions.
+- severity of the error ("Major" or "Minor"). 
+- reduction of score (between 0.5 and 5 given the severity of the error)
+
+Your evaluation output:
+"""
+
+llm = Llama(
+    model_path=hf_hub_download(
+        repo_id=os.environ.get("REPO_ID", "TIGER-Lab/TIGERScore-7B-GGUF"),
+        filename=os.environ.get("MODEL_FILE", "ggml-model-q4_0.gguf"),
+    ),
+    n_ctx=2048,
+    # n_gpu_layers=50, # change n_gpu_layers if you have more or less VRAM 
+) 
+
+def generate_text(input_context, generation_instruction, hypo_output, max_new_tokens=1024, temperature=0.7, top_p=1.0):
+    prompt_template = Template(TEMPLATE)
+    prompt = prompt_template.substitute(
+            generation_instruction=generation_instruction,
+            input_context=input_context,
+            hypothesis_output=hypo_output,
+        ).strip("\n ")
+    gen_params = {
+        "max_tokens": max_new_tokens,
+        "top_p": top_p,
+        "top_k": 40,
+        "temperature": temperature,
+        "frequency_penalty": 0.0,
+        "presence_penalty": 0.0,
+        "echo": False,
+        "stream": True,
+    }
+    outputs = llm(prompt, **gen_params)
+    temp=""
+    for out in outputs:
+        stream = copy.deepcopy(out)
+        temp += stream["choices"][0]["text"]
+        yield temp
         
 
 def get_examples(inst_textbox, input_textbox, hypo_output_textbox):
@@ -66,7 +106,7 @@ with gr.Blocks(theme='gradio/soft') as demo:
             minimum=256,
             maximum=1024,
             step=1,
-            value=512,
+            value=1024,
         )
         temperature = gr.Slider(
             label='Temperature of generation',
@@ -85,10 +125,10 @@ with gr.Blocks(theme='gradio/soft') as demo:
 
     gr.Markdown("## TIGERScore Outputs")
     evaluation_output_textbox = gr.Textbox(lines=4, label="Evaluation Output", placeholder="Evaluation output", show_label=True)
-
+    
 
     submit_button.click(
-        fn=submit_fn,
+        fn=generate_text,
         inputs=[input_textbox, inst_textbox, hypo_output_textbox, max_new_tokens, temperature, top_p],
         outputs=evaluation_output_textbox,
     )

requirements.txt

@@ -1,2 +1,5 @@
-git+https://github.com/TIGER-AI-Lab/TIGERScore.git
-datasets==2.14.5
+datasets==2.14.5
+torch
+transformers
+llama
+llama-cpp-python