it looks it is not compatible with other models downloaded from the same source (https://gpt4all.io/index.html)

[77bgh]

Hello, I just replaced the model path to use a different one, downloaded from https://gpt4all.io/index.html please note this model is in the same folder 'models' as ggml-gpt4all-j-v1.3-groovy.bin however with groovy (downloaded from https://gpt4all.io/index.html ) it works, as well all dependencies are installed accordingly with the required version

this is my .env,

PERSIST_DIRECTORY=db MODEL_TYPE=GPT4All MODEL_PATH=models/ggml-gpt4all-l13b-snoozy.bin EMBEDDINGS_MODEL_NAME=all-MiniLM-L6-v2 MODEL_N_CTX=1000 TARGET_SOURCE_CHUNKS=4

I tried as well with other models downloaded from the same source, the same error:

Using embedded DuckDB with persistence: data will be stored in: db Found model file. gptj_model_load: loading model from 'models/ggml-gpt4all-l13b-snoozy.bin' - please wait ... gptj_model_load: invalid model file 'models/ggml-gpt4all-l13b-snoozy.bin' (bad magic) GPT-J ERROR: failed to load model from models/ggml-gpt4all-l13b-snoozy.bin Enter a query:

[ProxyAyush]

got the same problem

[pseudotensor]

FYI Bad magic means different quantization version. I recommend, like localGPT and h2oGPT (https://github.com/h2oai/h2ogpt), using only latest llama.cpp based models, not GPT4All that are still on quantization 2. E.g. wizardlm or vicuna llama.cpp models are much better than gpt4all gpt-j etc.

[grigio]

@pseudotensor Which WizardLM did you try ? I tried with Wizard-Vicuna-7B-Uncensored.ggmlv3.q4_0.bin but I get Bad Magic

[77bgh]

it's working with different model "paraphrase-MiniLM-L6-v2" , looks faster

however if you ask him :"create in python a df with 2 columns: fist_name and last_name and populate it with 10 fake names, then print the results"

it will not do it as paraphrase-MiniLM-L6-v2 is different than all-MiniLM-L6-v2 , using all-MiniLM-L6-v2 it will do it, but it's longer the answer

1. "paraphrase-MiniLM-L6-v2": This model is based on the MiniLM architecture and has 6 layers. It is specifically designed for paraphrasing tasks, which involve generating alternative versions of a given text while preserving the original meaning. This model is fine-tuned on paraphrase data to optimize its performance for such tasks. It is a smaller version of the model compared to the "all-MiniLM" variants.

2. "all-MiniLM-L6-v2": This model, also based on the MiniLM architecture, consists of 6 layers. It is a general-purpose language model designed to handle various natural language processing tasks. It can be used for tasks such as text classification, sentiment analysis, question answering, and more. However, unlike the "paraphrase-MiniLM-L6-v2" model, it is not specifically optimized for paraphrasing.

3. "all-MiniLM-L12-v2": This model is similar to the "all-MiniLM-L6-v2" model in terms of being a general-purpose language model based on the MiniLM architecture. However, it is more powerful as it consists of 12 layers, which allows for deeper representations of language and potentially better performance on complex tasks. The additional layers enable the model to capture more intricate patterns and dependencies in the text, resulting in improved language understanding capabilities.

.env

PERSIST_DIRECTORY=db MODEL_TYPE=LlamaCpp MODEL_PATH=models/gpt4-x-alpaca-13b.ggmlv3.q5_1.bin EMBEDDINGS_MODEL_NAME=paraphrase-MiniLM-L6-v2 MODEL_N_CTX=1400 MODEL_N_BATCH=8 TARGET_SOURCE_CHUNKS=2