New version where the streaming streamlit app is working. Refactoring of the prompt mechanism. New, more efficient prompt with citations of the URL.
This commit is contained in:
parent
62559dcd0f
commit
ad9e7d93aa
1
.gitignore
vendored
1
.gitignore
vendored
@ -11,3 +11,4 @@ index_cera2/
|
||||
index_cera2_distiluse/
|
||||
__pycache__/
|
||||
chromadbtest/
|
||||
rag.log
|
||||
|
||||
2
PAD
2
PAD
@ -65,3 +65,5 @@ pip install langchain
|
||||
pip install -U sentence-transformers
|
||||
|
||||
pip install streamlit
|
||||
|
||||
python -m streamlit run app.py
|
||||
|
||||
26
app.py
26
app.py
@ -3,14 +3,16 @@ from rag import RAG
|
||||
import re
|
||||
import logging
|
||||
|
||||
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
||||
@st.cache_resource
|
||||
def init_rag():
|
||||
llm_model_path = '/Users/peportier/llm/a/a/zephyr-7b-beta.Q5_K_M.gguf'
|
||||
embed_model_name = 'intfloat/multilingual-e5-large'
|
||||
collection_name = 'cera'
|
||||
chromadb_path = './chromadb'
|
||||
rag = RAG(llm_model_path, embed_model_name, collection_name, chromadb_path)
|
||||
return rag
|
||||
|
||||
llm_model_path = '/Users/peportier/llm/a/a/zephyr-7b-beta.Q5_K_M.gguf'
|
||||
embed_model_name = 'intfloat/multilingual-e5-large'
|
||||
collection_name = 'cera'
|
||||
chromadb_path = './chromadb'
|
||||
|
||||
rag = RAG(llm_model_path, embed_model_name, collection_name, chromadb_path)
|
||||
rag = init_rag()
|
||||
|
||||
st.title("CERA Chat")
|
||||
|
||||
@ -37,10 +39,10 @@ if prompt := st.chat_input("Comment puis-je vous aider ?"):
|
||||
full_response += response
|
||||
message_placeholder.markdown(full_response + "▌")
|
||||
message_placeholder.markdown(full_response)
|
||||
url_pattern = r"URL :\n(https?://[^\s]+)"
|
||||
urls = re.findall(url_pattern, rag.chat_history[-1]['assistant'])
|
||||
markdown_urls = "\n".join([f"- {url}" for url in urls])
|
||||
logging.info(markdown_urls)
|
||||
#message_placeholder.markdown(markdown_urls)
|
||||
# url_pattern = r"URL :\n(https?://[^\s]+)"
|
||||
# urls = re.findall(url_pattern, rag.chat_history[-1]['assistant'])
|
||||
# markdown_urls = "\n".join([f"- {url}" for url in urls])
|
||||
# logging.info(f"URLs: \n{markdown_urls}")
|
||||
# message_placeholder.markdown(markdown_urls)
|
||||
|
||||
st.session_state.messages.append({"role": "assistant", "content": full_response})
|
||||
21
debug.py
Normal file
21
debug.py
Normal file
@ -0,0 +1,21 @@
|
||||
from rag import RAG
|
||||
|
||||
llm_model_path = '/Users/peportier/llm/a/a/zephyr-7b-beta.Q5_K_M.gguf'
|
||||
embed_model_name = 'intfloat/multilingual-e5-large'
|
||||
collection_name = 'cera'
|
||||
chromadb_path = './chromadb'
|
||||
|
||||
rag = RAG(llm_model_path, embed_model_name, collection_name, chromadb_path)
|
||||
|
||||
query1 = "Comment aider une entreprise qui rencontre des problèmes de trésorerie ?"
|
||||
ans1 = rag.chat(query1, stream=True)
|
||||
query2 = "Pouvez-vous m'en dire plus au sujet du deuxième point ?"
|
||||
ans2 = rag.chat(query2, stream=True)
|
||||
|
||||
# Queries:
|
||||
#
|
||||
# Lorsque mon client est en télétravail, quels sont les risques couverts par son assurance habitation ?
|
||||
#
|
||||
# Quel est le risque de perte attaché à la détention de Parts Sociales ?
|
||||
#
|
||||
# Comment procéder pour déclarer un sinistre habitation ? ou Comment procéder pour déclarer un sinistre Visa Premier ?
|
||||
231
rag.py
231
rag.py
@ -3,48 +3,103 @@ import chromadb
|
||||
from llama_cpp import Llama
|
||||
import copy
|
||||
import logging
|
||||
import re
|
||||
|
||||
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
||||
logging.basicConfig(filename='rag.log', level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
||||
|
||||
|
||||
class RAG:
|
||||
def __init__(self, llm_model_path, embed_model_name, collection_name, chromadb_path):
|
||||
logging.info('INIT')
|
||||
self.chat_history = []
|
||||
self.rag_system_prompt = """
|
||||
Vous êtes un assistant IA qui répond à la question posée par l'utilisateur en utilisant un contexte répertorié ci-dessous dans la rubrique Contexte.
|
||||
Le contexte est formé de passages exraits du site web commercial de la Caisse d'Epargne Rhône-Alpes, une banque française régionale.
|
||||
Votre réponse ne doit pas mentionner des informations déjà présentes dans l'historique de la conversation qui est répertorié ci-dessous dans la rubrique Historique.
|
||||
Vous fournissez avec soin des réponses précises, factuelles, réfléchies et nuancées, et vous êtes doué pour le raisonnement.
|
||||
Toutes les informations factuelles que vous utilisez pour répondre proviennent exclusivement du contexte.
|
||||
Si vous ne pouvez pas répondre à la question sur la base des éléments du contexte, dites simplement que vous ne savez pas, n'essayez pas d'inventer une réponse.
|
||||
Vos réponses doivent être brèves.
|
||||
Vos utilisateurs savent que vos réponses sont brèves et qu'elles ne mentionnent que les éléments du contexte, il n'est pas nécessaire de le leur rappeler.
|
||||
Vous gardez le rôle d'assistant et vous ne générez jamais le texte '<|user|>'.
|
||||
Vous rédigez vos réponses en français au format markdown sous forme d'une liste composée de 1 à 7 éléments au maximum.
|
||||
Voici le format que doit prendre votre réponse :
|
||||
```
|
||||
Voici des éléments de réponse :
|
||||
1. Elément de réponse.
|
||||
2. Elément de réponse.
|
||||
3. Elément de réponse.
|
||||
4. ...
|
||||
```
|
||||
self.tag_system = '<|system|>'
|
||||
self.tag_user = '<|user|>'
|
||||
self.tag_assistant = '<|assistant|>'
|
||||
self.tag_end = '</s>'
|
||||
self.rag_prompt = """
|
||||
{tag_system}
|
||||
Objectif
|
||||
========
|
||||
|
||||
----------------------------------------
|
||||
Historique :
|
||||
Vous êtes un assistant IA spécialiste des produits et services de la Caisse d'Epargne Rhône-Alpes, \
|
||||
une banque régionale française.
|
||||
Vous aidez un conseiller clientèle de la banque à mieux répondre aux besoins des clients.
|
||||
Vous fournissez avec soin des réponses précises et factuelles aux questions du conseiller.
|
||||
|
||||
Utilisation du contexte
|
||||
=======================
|
||||
|
||||
Vous répondez à la question posée par le conseiller en utilisant un contexte \
|
||||
formé de passages exraits du site web commercial de la banque.
|
||||
Votre réponse se base exclusivement sur les informations factuelles présentes dans le contexte.
|
||||
Si vous ne pouvez pas répondre à la question sur la base des éléments du contexte, \
|
||||
dites simplement que vous ne savez pas, n'essayez pas d'inventer une réponse.
|
||||
Voici le format d'un passage du contexte :
|
||||
```
|
||||
Titre :
|
||||
Le titre du passage
|
||||
|
||||
Catégorie :
|
||||
La catégorie du passage
|
||||
|
||||
URL :
|
||||
https://www.caisse-epargne.fr/rhone-alpes/url/du/passage
|
||||
|
||||
Contenu :
|
||||
Le contenu du passage
|
||||
```
|
||||
Vos réponses doivent toujours citer l'URL des passages utilisés. \
|
||||
Assurez-vous que l'URL citée correspond exactement à celle du passage. \
|
||||
Ne générez pas de nouvelles URLs. \
|
||||
Les conseillers sont encouragés à vérifier les URLs citées.
|
||||
|
||||
Format de réponse
|
||||
=================
|
||||
|
||||
Formulez chaque réponse sous forme de recommandations directes et concises, \
|
||||
en utilisant le langage et les termes présents dans le contexte.
|
||||
Citez l'URL en fin de réponse ou immédiatement après la recommandation pertinente.
|
||||
Vous rédigez votre réponse en français sous forme d'une liste d'informations \
|
||||
synthétiques extraites du contexte et qui seront utiles au conseiller.
|
||||
Vos utilisateurs savent qui vous êtes et quelles instructions vous avez reçues, \
|
||||
il n'est pas nécessaire de le leur rapeler.
|
||||
Voici le format que doit suivre votre réponse :
|
||||
```
|
||||
Voici des informations qui pourront aider votre client :
|
||||
|
||||
1. Utilisez [une solution spécifique du contexte] pour [traiter un aspect du problème]. Par exemple, [détail concret tiré du contexte]. Pour plus d'informations voir https://www.caisse-epargne.fr/rhone-alpes/url/du/passage
|
||||
|
||||
2. Considérez [une autre solution du contexte], qui est particulièrement adaptée pour [un autre aspect du problème]. Par exemple, [autre détail concret du contexte]. Pour plus d'informations voir https://www.caisse-epargne.fr/rhone-alpes/url/du/passage
|
||||
|
||||
```
|
||||
{tag_end}
|
||||
{history}
|
||||
----------------------------------------
|
||||
|
||||
{tag_user}
|
||||
Contexte :
|
||||
==========
|
||||
{context}
|
||||
|
||||
Question de l'utilisateur :
|
||||
===========================
|
||||
{query}
|
||||
{tag_end}
|
||||
{tag_assistant}
|
||||
Voici des informations qui pourront aider votre client :
|
||||
|
||||
1.
|
||||
"""
|
||||
self.query_reformulate_system_prompt = """
|
||||
self.query_reformulate_prompt = """
|
||||
{tag_system}
|
||||
Instructions :
|
||||
==============
|
||||
|
||||
Vous êtes un interprète conversationnel pour une conversation entre un utilisateur et \
|
||||
un assistant IA spécialiste des produits et services de la Caisse d'Epargne Rhône-Alpes, \
|
||||
une banque régionale française.
|
||||
L'utilisateur vous posera une question sans contexte. \
|
||||
L'utilisateur vous posera une question sans contexte.
|
||||
Vous devez reformuler la question pour prendre en compte le contexte de la conversation.
|
||||
Vous devez supposer que la question est liée aux produits et services de la Caisse d'Epargne Rhône-Alpes.
|
||||
Vous devez également consulter l'historique de la conversation ci-dessous lorsque vous reformulez la question. \
|
||||
Vous devez également consulter l'historique de la conversation ci-dessous lorsque vous reformulez la question.
|
||||
Par exemple, vous remplacerez les pronoms par les noms les plus probables dans l'historique de la conversation.
|
||||
Lorsque vous reformulez la question, accordez plus d'importance à la dernière question et \
|
||||
à la dernière réponse dans l'historique des conversations.
|
||||
@ -53,26 +108,36 @@ de sorte que l'échange le plus récent se trouve en haut de la page.
|
||||
Répondez en seulement une phrase avec la question reformulée.
|
||||
|
||||
Historique de la conversation :
|
||||
===============================
|
||||
|
||||
{history}
|
||||
{tag_end}
|
||||
{tag_user}
|
||||
Reformulez la question suivante : "{query}"
|
||||
{tag_end}
|
||||
{tag_assistant}
|
||||
Question reformulée : "
|
||||
"""
|
||||
|
||||
self.prefix_assistant_prompt = '1. '
|
||||
self.embed_model = SentenceTransformer(embed_model_name)
|
||||
self.chroma_client = chromadb.PersistentClient(path=chromadb_path)
|
||||
self.collection = self.chroma_client.get_collection(name=collection_name)
|
||||
self.llm = Llama(model_path=llm_model_path, n_gpu_layers=1, use_mlock=True, n_ctx=4096)
|
||||
|
||||
def answer(self, prompt, stream=False):
|
||||
def answer(self, prompt, stream):
|
||||
response = self.llm(prompt = prompt,
|
||||
temperature = 0.1,
|
||||
temperature = 0.7,
|
||||
mirostat_mode = 2,
|
||||
stream = stream,
|
||||
max_tokens = -1,
|
||||
stop = ['</s>', ' 8.', '\n\n', '<|user|>'])
|
||||
stop = [self.tag_end, self.tag_user])
|
||||
if stream:
|
||||
return response
|
||||
else: return response["choices"][0]["text"]
|
||||
|
||||
def query_collection(self, query, n_results=3):
|
||||
logging.info(f"query_collection / query: \n{query}")
|
||||
query = 'query: ' + query
|
||||
query_embedding = self.embed_model.encode(query, normalize_embeddings=True)
|
||||
query_embedding = query_embedding.tolist()
|
||||
@ -80,6 +145,13 @@ Historique de la conversation :
|
||||
query_embeddings=[query_embedding],
|
||||
n_results=n_results,
|
||||
)
|
||||
|
||||
ids_sources = ""
|
||||
for i in range(len(results["documents"][0])):
|
||||
id = results["ids"][0][i]
|
||||
ids_sources += id + " ; "
|
||||
logging.info(f"query_collection / sources: \n{ids_sources}")
|
||||
|
||||
return results
|
||||
|
||||
def format_passages(self, query_results):
|
||||
@ -107,70 +179,49 @@ Historique de la conversation :
|
||||
result = '\n'.join(result)
|
||||
return result
|
||||
|
||||
def format_rag_prompt(self, query, context="", history=""):
|
||||
|
||||
user_query = f"Question de l'utilisateur : \n{query}\n\n"
|
||||
assistant_answer = f"Réponse de l'assistant : \n 1. "
|
||||
self.chat_history.append({'user': user_query, 'assistant': assistant_answer})
|
||||
def answer_rag_prompt_streaming(self, prompt):
|
||||
logging.info(f"answer_rag_prompt_streaming: \n{prompt}")
|
||||
ans = self.answer(prompt, stream=True)
|
||||
|
||||
system_prompt = self.rag_system_prompt.format(history=history, context=context)
|
||||
|
||||
prompt = ""
|
||||
prompt = f"<|system|>\n{system_prompt.strip()}</s>\n"
|
||||
prompt += f"<|user|>\n{query}</s>\n"
|
||||
prompt += f"<|assistant|>\n Voici des éléments de réponse : \n 1. "
|
||||
|
||||
return prompt
|
||||
|
||||
def remove_references(self, text):
|
||||
motif = r"\(Passage \d+\)"
|
||||
res = re.sub(motif, '', text)
|
||||
return res
|
||||
yield self.prefix_assistant_prompt
|
||||
for item in ans:
|
||||
item = item["choices"][0]["text"]
|
||||
self.chat_history[-1]['assistant'] += item
|
||||
yield item
|
||||
|
||||
def answer_rag_prompt(self, query, query_results, stream=False):
|
||||
|
||||
query_context = self.format_passages(query_results)
|
||||
def answer_rag_prompt_non_streaming(self, prompt):
|
||||
logging.info(f"answer_rag_prompt_non_streaming: \n{prompt}")
|
||||
ans = self.answer(prompt, stream=False)
|
||||
|
||||
self.chat_history[-1]['assistant'] += ans
|
||||
ans = self.prefix_assistant_prompt + ans
|
||||
return ans
|
||||
|
||||
def prepare_prompt(self, query, query_results):
|
||||
context = self.format_passages(query_results)
|
||||
|
||||
history = ""
|
||||
for i in reversed(range(len(self.chat_history))):
|
||||
history += self.chat_history[i]["user"]
|
||||
history += self.remove_references(self.chat_history[i]["assistant"])
|
||||
history += "\n\n"
|
||||
|
||||
prompt = self.format_rag_prompt(query, query_context, history)
|
||||
for i in range(len(self.chat_history)):
|
||||
history += f"<|user|>\n{self.chat_history[i]['user']}</s>\n"
|
||||
history += f"<|assistant|>\n{self.chat_history[i]['assistant']}</s>\n"
|
||||
|
||||
logging.info(prompt)
|
||||
|
||||
ans = self.answer(prompt, stream)
|
||||
if stream:
|
||||
yield ' 1. '
|
||||
for item in ans:
|
||||
item = item["choices"][0]["text"]
|
||||
self.chat_history[-1]['assistant'] += item
|
||||
yield item
|
||||
else:
|
||||
self.chat_history[-1]['assistant'] += ans
|
||||
ans = '1. ' + ans
|
||||
return ans
|
||||
|
||||
def format_prompt_reformulate_query(self, query):
|
||||
system_prompt = self.query_reformulate_system_prompt
|
||||
|
||||
for i in reversed(range(len(self.chat_history))):
|
||||
system_prompt += self.chat_history[i]["user"]
|
||||
system_prompt += self.chat_history[i]["assistant"]
|
||||
self.chat_history.append({'user': query, 'assistant': self.prefix_assistant_prompt})
|
||||
|
||||
prompt = ""
|
||||
prompt = f"<|system|>\n{system_prompt.strip()}</s>\n"
|
||||
prompt += f"<|user|>\nPeux-tu reformuler la question suivante : \n \"{query}\"</s>\n"
|
||||
prompt += f"<|assistant|> Question reformulée : \n\""
|
||||
|
||||
return prompt
|
||||
return self.rag_prompt.format(history=history, query=query, context=context,
|
||||
tag_user=self.tag_user, tag_system=self.tag_system,
|
||||
tag_assistant=self.tag_assistant, tag_end=self.tag_end)
|
||||
|
||||
def reformulate_query(self, query):
|
||||
prompt = self.format_prompt_reformulate_query(query)
|
||||
logging.info(prompt)
|
||||
ans = self.answer(prompt)
|
||||
history = ""
|
||||
for i in reversed(range(len(self.chat_history))):
|
||||
history += f"Question de l'utilisateur :\n{self.chat_history[i]['user']}\n"
|
||||
history += f"Réponse de l'assistant :\n{self.chat_history[i]['assistant']}\n"
|
||||
|
||||
prompt = self.query_reformulate_prompt.format(history=history, query=query,
|
||||
tag_user=self.tag_user, tag_system=self.tag_system,
|
||||
tag_assistant=self.tag_assistant, tag_end=self.tag_end)
|
||||
logging.info(f"reformulate_query: \n{prompt}")
|
||||
ans = self.answer(prompt, stream=False)
|
||||
|
||||
last_quote_index = ans.rfind('"')
|
||||
if last_quote_index != -1:
|
||||
@ -183,11 +234,15 @@ Historique de la conversation :
|
||||
logging.info(f"La requête n'a pas pu être reformulée.")
|
||||
return query
|
||||
|
||||
def chat(self, query):
|
||||
def chat(self, query, stream=True):
|
||||
if len(self.chat_history) > 0:
|
||||
query = self.reformulate_query(query)
|
||||
query_results = self.query_collection(query)
|
||||
ans = self.answer_rag_prompt(query, query_results, stream=True)
|
||||
prompt = self.prepare_prompt(query, query_results)
|
||||
if stream:
|
||||
ans = self.answer_rag_prompt_streaming(prompt)
|
||||
else:
|
||||
ans = self.answer_rag_prompt_non_streaming(prompt)
|
||||
return ans
|
||||
|
||||
def reset_history(self):
|
||||
|
||||
@ -2,10 +2,21 @@
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"execution_count": 1,
|
||||
"id": "612c8bdb-83a8-4882-96a5-513ac7aedd7b",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stderr",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"/Users/peportier/miniforge3/envs/RAG_ENV/lib/python3.9/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
|
||||
" from .autonotebook import tqdm as notebook_tqdm\n",
|
||||
"/Users/peportier/miniforge3/envs/RAG_ENV/lib/python3.9/site-packages/transformers/utils/generic.py:441: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.\n",
|
||||
" _torch_pytree._register_pytree_node(\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"import importlib\n",
|
||||
"import rag\n",
|
||||
@ -16,10 +27,400 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"execution_count": 2,
|
||||
"id": "98130049-a4de-4532-8454-3df1a13094e7",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"metadata": {
|
||||
"collapsed": true,
|
||||
"jupyter": {
|
||||
"outputs_hidden": true
|
||||
}
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stderr",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"2024-01-04 09:15:27,599 - INFO - Load pretrained SentenceTransformer: intfloat/multilingual-e5-large\n",
|
||||
"/Users/peportier/miniforge3/envs/RAG_ENV/lib/python3.9/site-packages/transformers/utils/generic.py:309: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.\n",
|
||||
" _torch_pytree._register_pytree_node(\n",
|
||||
"2024-01-04 09:15:31,253 - INFO - Use pytorch device: cpu\n",
|
||||
"2024-01-04 09:15:31,257 - INFO - Anonymized telemetry enabled. See https://docs.trychroma.com/telemetry for more information.\n",
|
||||
"llama_model_loader: loaded meta data with 21 key-value pairs and 291 tensors from /Users/peportier/llm/a/a/zephyr-7b-beta.Q5_K_M.gguf (version GGUF V3 (latest))\n",
|
||||
"llama_model_loader: - tensor 0: token_embd.weight q5_K [ 4096, 32000, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 1: blk.0.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 2: blk.0.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 3: blk.0.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 4: blk.0.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 5: blk.0.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 6: blk.0.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 7: blk.0.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 8: blk.0.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 9: blk.0.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 10: blk.1.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 11: blk.1.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 12: blk.1.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 13: blk.1.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 14: blk.1.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 15: blk.1.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 16: blk.1.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 17: blk.1.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 18: blk.1.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 19: blk.2.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 20: blk.2.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 21: blk.2.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 22: blk.2.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 23: blk.2.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 24: blk.2.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 25: blk.2.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 26: blk.2.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 27: blk.2.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 28: blk.3.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 29: blk.3.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 30: blk.3.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 31: blk.3.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 32: blk.3.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 33: blk.3.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 34: blk.3.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 35: blk.3.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 36: blk.3.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 37: blk.4.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 38: blk.4.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 39: blk.4.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 40: blk.4.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 41: blk.4.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 42: blk.4.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 43: blk.4.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 44: blk.4.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 45: blk.4.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 46: blk.5.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 47: blk.5.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 48: blk.5.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 49: blk.5.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 50: blk.5.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 51: blk.5.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 52: blk.5.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 53: blk.5.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 54: blk.5.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 55: blk.6.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 56: blk.6.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 57: blk.6.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 58: blk.6.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 59: blk.6.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 60: blk.6.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 61: blk.6.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 62: blk.6.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 63: blk.6.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 64: blk.7.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 65: blk.7.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 66: blk.7.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 67: blk.7.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 68: blk.7.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 69: blk.7.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 70: blk.7.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 71: blk.7.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 72: blk.7.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 73: blk.8.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 74: blk.8.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 75: blk.8.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 76: blk.8.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 77: blk.10.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 78: blk.10.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 79: blk.10.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 80: blk.10.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 81: blk.10.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 82: blk.10.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 83: blk.10.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 84: blk.10.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 85: blk.10.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 86: blk.11.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 87: blk.11.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 88: blk.11.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 89: blk.11.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 90: blk.11.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 91: blk.11.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 92: blk.11.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 93: blk.11.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 94: blk.11.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 95: blk.12.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 96: blk.12.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 97: blk.12.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 98: blk.12.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 99: blk.12.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 100: blk.12.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 101: blk.8.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 102: blk.8.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 103: blk.8.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 104: blk.8.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 105: blk.8.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 106: blk.9.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 107: blk.9.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 108: blk.9.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 109: blk.9.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 110: blk.9.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 111: blk.9.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 112: blk.9.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 113: blk.9.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 114: blk.9.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 115: blk.12.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 116: blk.12.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 117: blk.12.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 118: blk.13.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 119: blk.13.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 120: blk.13.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 121: blk.13.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 122: blk.13.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 123: blk.13.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 124: blk.13.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 125: blk.13.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 126: blk.13.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 127: blk.14.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 128: blk.14.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 129: blk.14.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 130: blk.14.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 131: blk.14.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 132: blk.14.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 133: blk.14.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 134: blk.14.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 135: blk.14.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 136: blk.15.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 137: blk.15.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 138: blk.15.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 139: blk.15.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 140: blk.15.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 141: blk.15.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 142: blk.15.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 143: blk.15.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 144: blk.15.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 145: blk.16.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 146: blk.16.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 147: blk.16.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 148: blk.16.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 149: blk.16.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 150: blk.16.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 151: blk.16.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 152: blk.16.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 153: blk.16.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 154: blk.17.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 155: blk.17.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 156: blk.17.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 157: blk.17.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 158: blk.17.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 159: blk.17.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 160: blk.17.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 161: blk.17.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 162: blk.17.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 163: blk.18.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 164: blk.18.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 165: blk.18.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 166: blk.18.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 167: blk.18.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 168: blk.18.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 169: blk.18.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 170: blk.18.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 171: blk.18.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 172: blk.19.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 173: blk.19.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 174: blk.19.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 175: blk.19.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 176: blk.19.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 177: blk.19.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 178: blk.19.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 179: blk.19.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 180: blk.19.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 181: blk.20.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 182: blk.20.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 183: blk.20.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 184: blk.20.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 185: blk.20.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 186: blk.20.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 187: blk.20.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 188: blk.20.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 189: blk.20.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 190: blk.21.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 191: blk.21.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 192: blk.21.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 193: blk.21.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 194: blk.21.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 195: blk.21.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 196: blk.21.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 197: blk.21.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 198: blk.21.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 199: blk.22.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 200: blk.22.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 201: blk.22.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 202: blk.22.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 203: blk.22.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 204: blk.22.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 205: blk.22.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 206: blk.22.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 207: blk.22.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 208: blk.23.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 209: blk.23.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 210: blk.23.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 211: blk.23.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 212: blk.23.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 213: blk.23.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 214: blk.23.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 215: blk.23.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 216: blk.23.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 217: blk.24.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 218: blk.24.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 219: blk.24.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 220: blk.24.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 221: blk.24.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 222: blk.24.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 223: blk.24.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 224: blk.24.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 225: blk.24.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 226: blk.25.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 227: blk.25.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 228: blk.25.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 229: blk.25.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 230: blk.25.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 231: blk.25.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 232: blk.25.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 233: blk.25.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 234: blk.25.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 235: blk.26.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 236: blk.26.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 237: blk.26.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 238: blk.26.attn_v.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 239: blk.26.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 240: blk.26.ffn_down.weight q5_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 241: blk.26.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 242: blk.26.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 243: blk.26.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 244: blk.27.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 245: blk.27.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 246: blk.27.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 247: blk.27.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 248: blk.27.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 249: blk.27.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 250: blk.27.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 251: blk.27.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 252: blk.27.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 253: blk.28.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 254: blk.28.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 255: blk.28.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 256: blk.28.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 257: blk.28.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 258: blk.28.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 259: blk.28.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 260: blk.28.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 261: blk.28.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 262: blk.29.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 263: blk.29.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 264: blk.29.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 265: blk.29.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 266: blk.29.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 267: blk.29.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 268: blk.29.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 269: blk.29.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 270: blk.29.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 271: blk.30.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 272: blk.30.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 273: blk.30.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 274: blk.30.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 275: blk.30.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 276: blk.30.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 277: output.weight q6_K [ 4096, 32000, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 278: blk.30.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 279: blk.30.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 280: blk.30.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 281: blk.31.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 282: blk.31.ffn_down.weight q6_K [ 14336, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 283: blk.31.ffn_gate.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 284: blk.31.ffn_up.weight q5_K [ 4096, 14336, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 285: blk.31.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 286: blk.31.attn_k.weight q5_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 287: blk.31.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 288: blk.31.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 289: blk.31.attn_v.weight q6_K [ 4096, 1024, 1, 1 ]\n",
|
||||
"llama_model_loader: - tensor 290: output_norm.weight f32 [ 4096, 1, 1, 1 ]\n",
|
||||
"llama_model_loader: - kv 0: general.architecture str = llama\n",
|
||||
"llama_model_loader: - kv 1: general.name str = huggingfaceh4_zephyr-7b-beta\n",
|
||||
"llama_model_loader: - kv 2: llama.context_length u32 = 32768\n",
|
||||
"llama_model_loader: - kv 3: llama.embedding_length u32 = 4096\n",
|
||||
"llama_model_loader: - kv 4: llama.block_count u32 = 32\n",
|
||||
"llama_model_loader: - kv 5: llama.feed_forward_length u32 = 14336\n",
|
||||
"llama_model_loader: - kv 6: llama.rope.dimension_count u32 = 128\n",
|
||||
"llama_model_loader: - kv 7: llama.attention.head_count u32 = 32\n",
|
||||
"llama_model_loader: - kv 8: llama.attention.head_count_kv u32 = 8\n",
|
||||
"llama_model_loader: - kv 9: llama.attention.layer_norm_rms_epsilon f32 = 0.000010\n",
|
||||
"llama_model_loader: - kv 10: llama.rope.freq_base f32 = 10000.000000\n",
|
||||
"llama_model_loader: - kv 11: general.file_type u32 = 17\n",
|
||||
"llama_model_loader: - kv 12: tokenizer.ggml.model str = llama\n",
|
||||
"llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,32000] = [\"<unk>\", \"<s>\", \"</s>\", \"<0x00>\", \"<...\n",
|
||||
"llama_model_loader: - kv 14: tokenizer.ggml.scores arr[f32,32000] = [0.000000, 0.000000, 0.000000, 0.0000...\n",
|
||||
"llama_model_loader: - kv 15: tokenizer.ggml.token_type arr[i32,32000] = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...\n",
|
||||
"llama_model_loader: - kv 16: tokenizer.ggml.bos_token_id u32 = 1\n",
|
||||
"llama_model_loader: - kv 17: tokenizer.ggml.eos_token_id u32 = 2\n",
|
||||
"llama_model_loader: - kv 18: tokenizer.ggml.unknown_token_id u32 = 0\n",
|
||||
"llama_model_loader: - kv 19: tokenizer.ggml.padding_token_id u32 = 2\n",
|
||||
"llama_model_loader: - kv 20: general.quantization_version u32 = 2\n",
|
||||
"llama_model_loader: - type f32: 65 tensors\n",
|
||||
"llama_model_loader: - type q5_K: 193 tensors\n",
|
||||
"llama_model_loader: - type q6_K: 33 tensors\n",
|
||||
"llm_load_vocab: special tokens definition check successful ( 259/32000 ).\n",
|
||||
"llm_load_print_meta: format = GGUF V3 (latest)\n",
|
||||
"llm_load_print_meta: arch = llama\n",
|
||||
"llm_load_print_meta: vocab type = SPM\n",
|
||||
"llm_load_print_meta: n_vocab = 32000\n",
|
||||
"llm_load_print_meta: n_merges = 0\n",
|
||||
"llm_load_print_meta: n_ctx_train = 32768\n",
|
||||
"llm_load_print_meta: n_embd = 4096\n",
|
||||
"llm_load_print_meta: n_head = 32\n",
|
||||
"llm_load_print_meta: n_head_kv = 8\n",
|
||||
"llm_load_print_meta: n_layer = 32\n",
|
||||
"llm_load_print_meta: n_rot = 128\n",
|
||||
"llm_load_print_meta: n_gqa = 4\n",
|
||||
"llm_load_print_meta: f_norm_eps = 0.0e+00\n",
|
||||
"llm_load_print_meta: f_norm_rms_eps = 1.0e-05\n",
|
||||
"llm_load_print_meta: f_clamp_kqv = 0.0e+00\n",
|
||||
"llm_load_print_meta: f_max_alibi_bias = 0.0e+00\n",
|
||||
"llm_load_print_meta: n_ff = 14336\n",
|
||||
"llm_load_print_meta: rope scaling = linear\n",
|
||||
"llm_load_print_meta: freq_base_train = 10000.0\n",
|
||||
"llm_load_print_meta: freq_scale_train = 1\n",
|
||||
"llm_load_print_meta: n_yarn_orig_ctx = 32768\n",
|
||||
"llm_load_print_meta: rope_finetuned = unknown\n",
|
||||
"llm_load_print_meta: model type = 7B\n",
|
||||
"llm_load_print_meta: model ftype = mostly Q5_K - Medium\n",
|
||||
"llm_load_print_meta: model params = 7.24 B\n",
|
||||
"llm_load_print_meta: model size = 4.78 GiB (5.67 BPW) \n",
|
||||
"llm_load_print_meta: general.name = huggingfaceh4_zephyr-7b-beta\n",
|
||||
"llm_load_print_meta: BOS token = 1 '<s>'\n",
|
||||
"llm_load_print_meta: EOS token = 2 '</s>'\n",
|
||||
"llm_load_print_meta: UNK token = 0 '<unk>'\n",
|
||||
"llm_load_print_meta: PAD token = 2 '</s>'\n",
|
||||
"llm_load_print_meta: LF token = 13 '<0x0A>'\n",
|
||||
"llm_load_tensors: ggml ctx size = 0.11 MiB\n",
|
||||
"llm_load_tensors: mem required = 4893.10 MiB\n",
|
||||
"...................................................................................................\n",
|
||||
"llama_new_context_with_model: n_ctx = 4096\n",
|
||||
"llama_new_context_with_model: freq_base = 10000.0\n",
|
||||
"llama_new_context_with_model: freq_scale = 1\n",
|
||||
"llama_new_context_with_model: kv self size = 512.00 MiB\n",
|
||||
"llama_build_graph: non-view tensors processed: 740/740\n",
|
||||
"ggml_metal_init: allocating\n",
|
||||
"ggml_metal_init: found device: Apple M2 Max\n",
|
||||
"ggml_metal_init: picking default device: Apple M2 Max\n",
|
||||
"ggml_metal_init: default.metallib not found, loading from source\n",
|
||||
"ggml_metal_init: loading '/Users/peportier/miniforge3/envs/RAG_ENV/lib/python3.9/site-packages/llama_cpp/ggml-metal.metal'\n",
|
||||
"ggml_metal_init: GPU name: Apple M2 Max\n",
|
||||
"ggml_metal_init: GPU family: MTLGPUFamilyApple8 (1008)\n",
|
||||
"ggml_metal_init: hasUnifiedMemory = true\n",
|
||||
"ggml_metal_init: recommendedMaxWorkingSetSize = 49152.00 MiB\n",
|
||||
"ggml_metal_init: maxTransferRate = built-in GPU\n",
|
||||
"llama_new_context_with_model: compute buffer total size = 291.07 MiB\n",
|
||||
"llama_new_context_with_model: max tensor size = 102.54 MiB\n",
|
||||
"ggml_metal_add_buffer: allocated 'data ' buffer, size = 4893.70 MiB, ( 4894.33 / 49152.00)\n",
|
||||
"ggml_metal_add_buffer: allocated 'kv ' buffer, size = 512.02 MiB, ( 5406.34 / 49152.00)\n",
|
||||
"ggml_metal_add_buffer: allocated 'alloc ' buffer, size = 288.02 MiB, ( 5694.36 / 49152.00)\n",
|
||||
"AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | \n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"llm_model_path = '/Users/peportier/llm/a/a/zephyr-7b-beta.Q5_K_M.gguf'\n",
|
||||
"embed_model_name = 'intfloat/multilingual-e5-large'\n",
|
||||
@ -29,6 +430,35 @@
|
||||
"rag = RAG(llm_model_path, embed_model_name, collection_name, chromadb_path)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 3,
|
||||
"id": "b12ed9e5-cacc-4f9b-a6b9-38ccda00764f",
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stderr",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Batches: 100%|████████████████████████████████████| 1/1 [00:00<00:00, 2.19it/s]\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"query1 = \"Comment la Caisse d'Epargne Rhône-Alpes peut-elle aider une entreprise qui rencontre des problèmes de trésorerie ?\"\n",
|
||||
"res1 = rag.chat(query1, stream=False)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "086aef56-223f-4d3b-a1f0-9d251095e9f9",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"res1"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
|
||||
Loading…
Reference in New Issue
Block a user