FiveTech Software tech support forums

Marzo-Abril 2024 ================ * Nuevo: Soporte para Inteligencia Artificial local de LLM (modelos grandes de lenguaje) usando llama.cpp que ha sido incluido en FWH. Por favor, revisa samples\llm.prg. Para probarlo tienes que descargar tinyLlama.gguf desde aquí: ...

March-April 2024 ================ * New: Support for local AI LLM (large language models) using llama.cpp that has been included in FWH. Please review samples\llm.prg. In order to test it you have to download tinyLlama.gguf from here: https://huggingface.co/fivetech/tinyLlama/blob/main/tinyLlama.gguf. ...

MMLU (Masked Multi-Layer Unified) Supongamos que tenemos el siguiente texto: "The capital of France is [MASK]." Un modelo de lenguaje entrenado con MMLU intentaría predecir la palabra que falta en el texto. Por ejemplo, si el modelo predice que la palabra es "Paris", se consider...

Edited the first post of this topic with the right instructions:

viewtopic.php?p=266364&sid=34d610603696853e410ee75921e1424b#p266364

From posts.dbf and posts.fpt now we generate the dataset.json to be trained. We are using just 20 different topics, so the dataset is not too large and we can do quicker tests with it: dataset.prg #include "FiveWin.ch"request dbfcdxfunction Main()    local aPosts := {}, n...

... se queda tan ancha Ahora el gran paradigma es como enseñar a la IA de una empresa sobre algo en especial y que tipo de modelo de lenguaje como el LLM poder usar, y leo, me comentan , me explican,… que están obsesionados en volcar grandes cantidades de terabytes de info para que la IA pueda relacionar. ...

Here you have run.py to test the model: run.py from transformers import GPT2LMHeadModel, GPT2Tokenizerimport torch# Load the fine-tuned GPT-2 model and tokenizerfine_tuned_model_path = "./fine-tuned-model"model = GPT2LMHeadModel.from_pretrained(fine_tuned_model_path)tokenizer = GPT...

Dear Anton,

many thanks for your help!

I am reviewing the results

It's been two days since I've had anything to do with Python! DESKTOP PC with Windows 11 pro: Processor: Intel(R) Core(TM) i9-10920X CPU @ 3.50GHz, 3504 MHz, 12 core(s), 24 logical processor(s) Data carrier: Samsung SSD 970 EVO Plus 1TB 64GB RAM NVIDIA RTX 2080TI First attempt, without GPU support (...

Dear Anton,

many thanks for your so valuable feedback!

Hello Antonio, From the looks of it, the training doesn't just happen in one round: First round generated a 715MB file "cached_lm_GPT2Tokenizer_128_fivetech_forums_20231222.sql" in 16min. The second round is actually the training and I'm now at 0% after 47 minutes and the open duration is ...

Dear Anton,

> result: ready in 16min maybe with some errors - error messages can't be saved there!

Can you please send me te generated "fine-tuned-model" files using wormhole app ?

https://wormhole.app/

many thanks!

installed python 3.11.7
pip install torch
pip install transformers

python train.py

result: ready in 16min maybe with some errors - error messages can't be saved there!

if i run train.py again then comes terrible | 213/1628301 [05:57<1293:44:33, 2.86s/it]

can't wait so long

I think i should downgrade python to 3.11 https://stackoverflow.com/questions/77225812/is-there-a-way-to-install-pytorch-on-python-3-12-0 here is the output with conda ... C:\fwh\AI>conda install pytorch==1.10.0 torchvision==0.11.2 torchaudio==0.10.0 -c pytorch Channels: - pytorch - defau...

python is 3.12 pip is uptodate antivirus is stopped ? C:\fwh\AI>pip install pytorch Collecting pytorch Using cached pytorch-1.0.2.tar.gz (689 bytes) Installing build dependencies ... done Getting requirements to build wheel ... done Preparing metadata (pyproject.toml) ... done Building wheels for...