this should activate lfs file support

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+deps/intel_npu_acceleration_library-1.3.0-cp310-cp310-win_amd64.whl filter=lfs diff=lfs merge=lfs -text

app.py

@@ -16,43 +16,9 @@ user_msg = "Please describe what you need to do. To get the best results try to
 def check_for_missing_answers(parsed_questions: dict[int, str]):
     return [k for k in parsed_questions if parsed_questions[k] is None]
 
-
-def ask_again(missing_questions: list[int], user_questions: list[str], parsed_questions: dict[int, str]):
-    for id_ in missing_questions:
-        new_answer = input(f"I could not find the answer to this question: {user_questions[id_].lower()}")
-        parsed_questions[id_] = new_answer
-
-
-def use_command_line():
-    prompts_manager = PromptsManager()
-    user_prompt = input(f"{user_msg}\n{'\n'.join(prompts_manager.questions)}\n\n>")
-
-    repository = get_repository("intel_npu", Model("meta-llama/Meta-Llama-3-8B-Instruct",
-                                                   ModelRoles("system", "user", "assistant")),
-                                prompts_manager.system_prompt, Path("llm_log.txt"))
-    repository.init()
-    # repository.send_prompt(prompts_manager.ingest_user_answers(user_prompt))
-    answer = repository.send_prompt(prompts_manager.verify_user_input_prompt(user_prompt))
-    answers = LlmParser.parse_verification_prompt_answers(answer['content'])
-
-    missing_answers = check_for_missing_answers(answers)
-    while missing_answers:
-        ask_again(missing_answers, prompts_manager.questions, answers)
-        missing_answers = check_for_missing_answers(answers)
-    answer = repository.send_prompt(prompts_manager.get_work_category(answers[1]))
-    categories = LlmParser.parse_get_categories_answer(answer['content'])
-
-    form_data = build_form_data_from_answers(answers, categories, f"{Path(__file__, "..", "signature.png")}")
-    write_pdf_form(form_data, Path("signed_form1.pdf"))
-
-
-def update_answer (answers, missing_answer):
-    answers[missing_answer] = getattr(st.session_state, f"ma_{missing_answer}")
-
-
 def use_streamlit():
     pm = PromptsManager()
-    help_ = f"{user_msg}\n\n{'\n'.join(pm.questions)}"
+    help_ = f"{user_msg}\n\n" + '\n'.join(pm.questions)
     repository = get_repository("ondemand", Model("ondemand-gpt-3.5-turbo", ModelRoles("system", "user", "assistant")))
     if not st.session_state.get("step"):
         with st.form("Please describe your request"):
@@ -100,7 +66,7 @@ def use_streamlit():
 
         with st.status("categories found, creating PDF form"):
 
-            form_filename = f"{st.session_state["session_id"]}_form.pdf"
+            form_filename = f"{st.session_state['session_id']}_form.pdf"
             st.session_state["form_filename"] = form_filename
             form_data = build_form_data_from_answers(st.session_state["answers"], categories,
                                                      st.session_state.get("signature"))
@@ -116,4 +82,3 @@ def use_streamlit():
 
 
 use_streamlit()
-#use_command_line()

repository/intel_npu.py

@@ -1,8 +1,8 @@
 import json
 from pathlib import Path
 
-# from intel_npu_acceleration_library import NPUModelForCausalLM, int4
-# from intel_npu_acceleration_library.compiler import CompilerConfig
+from intel_npu_acceleration_library import NPUModelForCausalLM, int4
+from intel_npu_acceleration_library.compiler import CompilerConfig
 from transformers import AutoTokenizer
 
 from repository.repository_abc import Repository, Model
@@ -25,35 +25,34 @@ class IntelNpuRepository(Repository):
         return self.message_history
 
     def init(self):
-        pass
-        # compiler_conf = CompilerConfig(dtype=int4)
-        # self.model = NPUModelForCausalLM.from_pretrained(self.model_info.name, use_cache=True, config=compiler_conf,
-        #                                                  export=True, temperature=0).eval()
-        # self.tokenizer = AutoTokenizer.from_pretrained(self.model_info.name)
-        # self.terminators = [self.tokenizer.eos_token_id, self.tokenizer.convert_tokens_to_ids("<|eot_id|>")]
+        compiler_conf = CompilerConfig(dtype=int4)
+        self.model = NPUModelForCausalLM.from_pretrained(self.model_info.name, use_cache=True, config=compiler_conf,
+                                                          export=True, temperature=0).eval()
+        self.tokenizer = AutoTokenizer.from_pretrained(self.model_info.name)
+        self.terminators = [self.tokenizer.eos_token_id, self.tokenizer.convert_tokens_to_ids("<|eot_id|>")]
 
     def send_prompt(self, prompt: str, add_to_history: bool = True) -> dict[str, str]:
         pass
-        # print("prompt to be sent: " + prompt)
-        # user_prompt = {"role": self.model_info.roles.user_role, "content": prompt}
-        # if self.log_to_file:
-        #     with open(self.log_to_file, "a+") as log_file:
-        #         log_file.write(json.dumps(user_prompt, indent=2))
-        #         log_file.write("\n")
-        # self.get_message_history().append(user_prompt)
-        # input_ids = (self.tokenizer.apply_chat_template(self.get_message_history(), add_generation_prompt=True,
-        #                                                 return_tensors="pt")
-        #              .to(self.model.device))
-        # outputs = self.model.generate(input_ids, eos_token_id=self.terminators, do_sample=True, max_new_tokens=2000, cache_position=None)
-        # generated_token_array = outputs[0][len(input_ids[0]):]
-        # generated_tokens = "".join(self.tokenizer.batch_decode(generated_token_array, skip_special_tokens=True))
-        # answer = {"role": self.get_model_info().roles.ai_role, "content": generated_tokens}
-        # if self.log_to_file:
-        #     with open(self.log_to_file, "a+") as log_file:
-        #         log_file.write(json.dumps(answer, indent=2))
-        #         log_file.write("\n")
-        # if add_to_history:
-        #     self.message_history.append(answer)
-        # else:
-        #     self.message_history.pop()
-        # return answer
+        print("prompt to be sent: " + prompt)
+        user_prompt = {"role": self.model_info.roles.user_role, "content": prompt}
+        if self.log_to_file:
+            with open(self.log_to_file, "a+") as log_file:
+                log_file.write(json.dumps(user_prompt, indent=2))
+                log_file.write("\n")
+        self.get_message_history().append(user_prompt)
+        input_ids = (self.tokenizer.apply_chat_template(self.get_message_history(), add_generation_prompt=True,
+                                                        return_tensors="pt")
+                     .to(self.model.device))
+        outputs = self.model.generate(input_ids, eos_token_id=self.terminators, do_sample=True, max_new_tokens=2000, cache_position=None)
+        generated_token_array = outputs[0][len(input_ids[0]):]
+        generated_tokens = "".join(self.tokenizer.batch_decode(generated_token_array, skip_special_tokens=True))
+        answer = {"role": self.get_model_info().roles.ai_role, "content": generated_tokens}
+        if self.log_to_file:
+            with open(self.log_to_file, "a+") as log_file:
+                log_file.write(json.dumps(answer, indent=2))
+                log_file.write("\n")
+        if add_to_history:
+            self.message_history.append(answer)
+        else:
+            self.message_history.pop()
+        return answer

requirements.txt

@@ -1,4 +1,5 @@
 PyPDFForm
 ollama
 transformers
-# intel-npu-acceleration-library
+streamlit
+deps/intel_npu_acceleration_library-1.3.0-cp310-cp310-win_amd64.whl