From 8787780835d1699d8ce2da0680ffb00575f9a99d Mon Sep 17 00:00:00 2001
From: Aaron Giner <aaron.giner@student.tugraz.at>
Date: Tue, 23 Apr 2024 08:12:24 +0200
Subject: [PATCH] some cleanup and datetime parser fixes
---
python/llm-server/model.py | 10 +++----
python/llm-server/parse_dt.py | 15 ++++------
python/llm-server/queries.py | 56 ++++++++++++++++++++---------------
python/llm-server/util.py | 7 ++---
4 files changed, 45 insertions(+), 43 deletions(-)
diff --git a/python/llm-server/model.py b/python/llm-server/model.py
index 51e7dde..efbcf93 100644
--- a/python/llm-server/model.py
+++ b/python/llm-server/model.py
@@ -21,14 +21,14 @@ class Model:
if not self.use_chat_gpt:
if use_local_model:
- self.llm = util.load_model(model=model_name, ctx_size=self.ctx_size, temperature=temperature)
+ self.llm = util.load_model(model=model_name, ctx_size=self.ctx_size, temperature=self.temperature)
else:
- model_path = hf_hub_download(self.model_name, self.model_file, token=api_key)
- self.llm = util.load_model(model=model_path, ctx_size=self.ctx_size, temperature=temperature)
+ model_path = hf_hub_download(self.model_name, self.model_file, token=self.api_key)
+ self.llm = util.load_model(model=model_path, ctx_size=self.ctx_size, temperature=self.temperature)
- self.tokenizer = util.load_tokenizer(model=self.model_tok, token=util.LLAMA_API_KEY)
+ self.tokenizer = util.load_tokenizer(model=self.model_tok, token=self.api_key)
- def query(self, messages, debug=False):
+ def query(self, messages, debug=True):
if self.use_chat_gpt:
return self.query_openai(messages=messages)
diff --git a/python/llm-server/parse_dt.py b/python/llm-server/parse_dt.py
index 1f7f201..e2d618f 100644
--- a/python/llm-server/parse_dt.py
+++ b/python/llm-server/parse_dt.py
@@ -7,27 +7,24 @@ import parsedatetime as pdt
def replace_all(dt, statements):
cal = pdt.Calendar()
dt = dateutil.parser.parse(dt)
- dt_now = datetime.now()
replaced = []
for statement in statements:
- print(statement)
parsed_nlp = cal.nlp(statement, dt)
- parsed_nlp_now = cal.nlp(statement, dt_now)
if parsed_nlp is None or len(parsed_nlp) == 0:
replaced.append(statement)
continue
- parsed_nlp = parsed_nlp[0]
- # if the 2 times are different, that very likely means that the time was not mentioned in the statement
- if parsed_nlp[0].time() == parsed_nlp_now[0][0].time():
- include_time = True
- else:
+ # check datetimeflags to see if a date or datetime was parsed
+ print(parsed_nlp[0][1])
+ if parsed_nlp[0][1] == 1:
include_time = False
+ else:
+ include_time = True
dt_format = "%A, %B %d, %Y" + (", %I:%M %p" if include_time else "")
- statement = statement.replace(parsed_nlp[-1], "on " + str(parsed_nlp[0].strftime(dt_format)))
+ statement = statement.replace(parsed_nlp[0][-1], "on " + str(parsed_nlp[0][0].strftime(dt_format)))
replaced.append(statement)
return replaced
diff --git a/python/llm-server/queries.py b/python/llm-server/queries.py
index 0273b58..cec279b 100644
--- a/python/llm-server/queries.py
+++ b/python/llm-server/queries.py
@@ -18,32 +18,40 @@ from datetime import datetime, timedelta
def parse_request(request):
request = json.loads(request)
- requestType = request["type"]
+ requestType = request["Type"]
+
+ model_info = request["Model"]
+ model_repo = model_info["ModelRepo"]
+ model_file = model_info["ModelFile"]
+ model_tok_repo = model_info["ModelTokRepo"]
+ api_key = model_info["APIKey"]
+ use_chat_gpt = model_info["UseChatGPT"]
+
+ llm = model.Model(model_repo, model_file, model_tok_repo, 0.75, 4096, api_key, use_chat_gpt)
if requestType == "chat":
- return query_chat(request, model.Model(util.LLAMA_NAME, util.LLAMA_FILE, util.LLAMA_TOK, 0.75, 4096,
- "sk-proj-aUDdsiCXHDwoHewZFL9AT3BlbkFJIkKZEYaMi5AGEBDbW2zv", use_chat_gpt=True))
+ return query_chat(request, llm)
elif requestType == "chat_summary":
- return query_chat_summary(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return query_chat_summary(request, llm)
elif requestType == "chat_extract_plan":
- return query_chat_extract_plan(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return query_chat_extract_plan(request, llm)
elif requestType == "reflection":
- return query_reflection(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return query_reflection(request, llm)
elif requestType == "poignancy":
- return query_poignancy(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return query_poignancy(request, llm)
elif requestType == "context":
- return generate_context(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return generate_context(request, llm)
elif requestType == "plan_day":
- return query_plan_day(request, model.Model(util.MISTRAL_GGUF, util.MISTRAL_FILE, util.MISTRAL_TOK, 0, 4096))
+ return query_plan_day(request, llm)
return "ERROR"
def query_chat(request, llm: model.Model):
- parameters = request["data"]
+ parameters = request["Data"]
chat = parameters["chat"].split("~")
- memories = request["memories"]
+ memories = request["Memories"]
memory_util.filter_memories(memories, parameters["user"] + ";" + chat[-1], parameters["datetime"])
parameters["memories"] = memory_util.memories_to_string(memories[:10], False)
@@ -67,8 +75,8 @@ def query_chat(request, llm: model.Model):
def query_reflection(request, llm: model.Model):
- parameters = request["data"]
- memories = request["memories"]
+ parameters = request["Data"]
+ memories = request["Memories"]
parameters["memories"] = memory_util.memories_to_string(memories, include_nodeId=True)
messages = [
@@ -84,7 +92,7 @@ def query_reflection(request, llm: model.Model):
def query_poignancy(request, llm: model.Model):
- parameters = request["data"]
+ parameters = request["Data"]
messages = [
{"role": "user",
"content": util.load_template("poignancy").format(**parameters)},
@@ -141,7 +149,7 @@ def deduplicate_sim(statements):
def query_chat_summary_extract(template, request, llm: model.Model, ctx_window=5):
- parameters = request["data"]
+ parameters = request["Data"]
chat = parameters["conversation"]
lines = chat.split("\n")
@@ -180,7 +188,7 @@ def query_chat_summary_extract(template, request, llm: model.Model, ctx_window=5
def query_chat_summary(request, llm: model.Model):
- parameters = request["data"]
+ parameters = request["Data"]
messages_summary = [
{"role": "user",
"content": util.load_template("chat_summary_single").format(**parameters)},
@@ -207,7 +215,7 @@ def query_chat_summary(request, llm: model.Model):
def query_chat_extract_plan(request, llm: model.Model):
- parameters = request["data"]
+ parameters = request["Data"]
# alternative 1
plans = query_chat_summary_extract("chat_extract_plan", request, llm, 10)
@@ -278,8 +286,8 @@ def get_calendar():
def generate_context(request, llm: model.Model):
- parameters = request["data"]
- memories = request["memories"]
+ parameters = request["Data"]
+ memories = request["Memories"]
memories.sort(key=lambda x: x["HrsSinceCreation"], reverse=True)
# if the agent has no memory associated with the user, then they have never had conversation
@@ -303,8 +311,8 @@ def generate_context(request, llm: model.Model):
def query_relationship(request, llm: model.Model):
- parameters = request["data"]
- memories = request["memories"]
+ parameters = request["Data"]
+ memories = request["Memories"]
memories_str = memory_util.memories_to_string(memories, include_date_created=True)
@@ -319,7 +327,7 @@ def query_relationship(request, llm: model.Model):
def query_agent_action(request, llm: model.Model):
- parameters = request["data"]
+ parameters = request["Data"]
messages = [
{"role": "user",
"content": util.load_template("agent_action").format(**parameters)},
@@ -331,8 +339,8 @@ def query_agent_action(request, llm: model.Model):
def query_plan_day(request, llm: model.Model):
- parameters = request["data"]
- memories = request["memories"]
+ parameters = request["Data"]
+ memories = request["Memories"]
memories_str = memory_util.memories_to_string(memories)
diff --git a/python/llm-server/util.py b/python/llm-server/util.py
index 2991ca3..ed48925 100644
--- a/python/llm-server/util.py
+++ b/python/llm-server/util.py
@@ -21,17 +21,14 @@ MISTRAL_FILE = "mistral-7b-instruct-v0.1.Q4_K_M.gguf"
LLAMA_TOK = "meta-llama/Llama-2-7b-chat-hf"
LLAMA_API_KEY = "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"
-LLAMA_NAME = "TheBloke/Llama-2-7B-GGUF"
+LLAMA_NAME = "TheBloke/Llama-2-7B-Chat-GGUF"
LLAMA_FILE = "llama-2-7b.Q4_K_M.gguf"
KW_EN_CORE = "X:/LLM Models/nlm/en_core_web_sm-3.7.1-py3-none-any.whl"
def load_tokenizer(model, token=""):
- if "llama" in model:
- return AutoTokenizer.from_pretrained(model, token=token)
- else:
- return AutoTokenizer.from_pretrained(model)
+ return AutoTokenizer.from_pretrained(model, token=token)
def load_model(model, ctx_size, temperature):
--
GitLab