Quantization made by Richard Erkhov.

Meta-Llama-3-8B-InitializedEmbeds - GGUF

Model creator: https://huggingface.co./chargoddard/
Original model: https://huggingface.co./chargoddard/Meta-Llama-3-8B-InitializedEmbeds/

Name	Quant method	Size
Meta-Llama-3-8B-InitializedEmbeds.Q2_K.gguf	Q2_K	2.96GB
Meta-Llama-3-8B-InitializedEmbeds.IQ3_XS.gguf	IQ3_XS	3.28GB
Meta-Llama-3-8B-InitializedEmbeds.IQ3_S.gguf	IQ3_S	3.43GB
Meta-Llama-3-8B-InitializedEmbeds.Q3_K_S.gguf	Q3_K_S	3.41GB
Meta-Llama-3-8B-InitializedEmbeds.IQ3_M.gguf	IQ3_M	3.52GB
Meta-Llama-3-8B-InitializedEmbeds.Q3_K.gguf	Q3_K	3.74GB
Meta-Llama-3-8B-InitializedEmbeds.Q3_K_M.gguf	Q3_K_M	3.74GB
Meta-Llama-3-8B-InitializedEmbeds.Q3_K_L.gguf	Q3_K_L	4.03GB
Meta-Llama-3-8B-InitializedEmbeds.IQ4_XS.gguf	IQ4_XS	4.18GB
Meta-Llama-3-8B-InitializedEmbeds.Q4_0.gguf	Q4_0	4.34GB
Meta-Llama-3-8B-InitializedEmbeds.IQ4_NL.gguf	IQ4_NL	4.38GB
Meta-Llama-3-8B-InitializedEmbeds.Q4_K_S.gguf	Q4_K_S	4.37GB
Meta-Llama-3-8B-InitializedEmbeds.Q4_K.gguf	Q4_K	4.58GB
Meta-Llama-3-8B-InitializedEmbeds.Q4_K_M.gguf	Q4_K_M	4.58GB
Meta-Llama-3-8B-InitializedEmbeds.Q4_1.gguf	Q4_1	4.78GB
Meta-Llama-3-8B-InitializedEmbeds.Q5_0.gguf	Q5_0	5.21GB
Meta-Llama-3-8B-InitializedEmbeds.Q5_K_S.gguf	Q5_K_S	5.21GB
Meta-Llama-3-8B-InitializedEmbeds.Q5_K.gguf	Q5_K	5.34GB
Meta-Llama-3-8B-InitializedEmbeds.Q5_K_M.gguf	Q5_K_M	5.34GB
Meta-Llama-3-8B-InitializedEmbeds.Q5_1.gguf	Q5_1	5.65GB
Meta-Llama-3-8B-InitializedEmbeds.Q6_K.gguf	Q6_K	6.14GB
Meta-Llama-3-8B-InitializedEmbeds.Q8_0.gguf	Q8_0	7.95GB

Original model description:

base_model:

NousResearch/Meta-Llama-3-8B-Instruct
NousResearch/Meta-Llama-3-8B library_name: transformers tags:
mergekit
merge

Meta-Llama-3-8B-InitializedEmbeds

This is just Llama-3-8B with the embeddings for special tokens copied from the Instruct version. Should behave pretty much identically to the base model, but with less glossolalia when it encounters <|start_header_id|> and the like.

I'm using this as a base to fine tune. Having these embeddings reasonable instead of randomly initialized should give a smoother start.

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the linear merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

merge_method: linear
dtype: float32
out_dtype: bfloat16
models:
  - model: NousResearch/Meta-Llama-3-8B
    parameters:
      weight: 1.0
  - model: NousResearch/Meta-Llama-3-8B-Instruct
    parameters:
      weight: 0.0
tokenizer:
  source: NousResearch/Meta-Llama-3-8B-Instruct
  tokens:
    <|start_header_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|end_header_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|eot_id|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|end_of_text|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true
    <|begin_of_text|>:
      source: NousResearch/Meta-Llama-3-8B-Instruct
      force: true