Instructions to use macadeliccc/magistrate-3.2-3b-it-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="macadeliccc/magistrate-3.2-3b-it-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("macadeliccc/magistrate-3.2-3b-it-GGUF", dtype="auto")

llama-cpp-python

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="macadeliccc/magistrate-3.2-3b-it-GGUF",
	filename="magistrate-3.2-3b-it.Q2_K.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

Use Docker

docker model run hf.co/macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "macadeliccc/magistrate-3.2-3b-it-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "macadeliccc/magistrate-3.2-3b-it-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

SGLang

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "macadeliccc/magistrate-3.2-3b-it-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "macadeliccc/magistrate-3.2-3b-it-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "macadeliccc/magistrate-3.2-3b-it-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "macadeliccc/magistrate-3.2-3b-it-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use macadeliccc/magistrate-3.2-3b-it-GGUF with Ollama:
```
ollama run hf.co/macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
```

Unsloth Studio new

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for macadeliccc/magistrate-3.2-3b-it-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for macadeliccc/magistrate-3.2-3b-it-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for macadeliccc/magistrate-3.2-3b-it-GGUF to start chatting

Docker Model Runner
How to use macadeliccc/magistrate-3.2-3b-it-GGUF with Docker Model Runner:
```
docker model run hf.co/macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M
```

Lemonade

How to use macadeliccc/magistrate-3.2-3b-it-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull macadeliccc/magistrate-3.2-3b-it-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.magistrate-3.2-3b-it-GGUF-Q4_K_M

List all available models

lemonade list

magistrate-3.2-3b-it

This model is a fine-tuned version of macadeliccc/magistrate-3.2-3b-base on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.8067

See axolotl config

axolotl version: 0.4.1

base_model: macadeliccc/magistrate-3.2-3b-base
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/hermes-2.5.jsonl
  # - path: json 
  #   type: sharegpt
  #   conversation: chatml
  #   data_files: train/financial_instructions_cleaned_2.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/glaive-function-calling-5k.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/func-calling-singleturn.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/func-calling.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/json-mode-agentic.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/json-mode-singleturn.json
  - path: json 
    type: sharegpt
    conversation: chatml
    data_files: train/reasoning_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/systemchat_2_0_small.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/303_creative_llc_v__elenis_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/abitron_austria_gmbh_v__hetronic_international__inc__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/acheson_hotels__llc_v__laufer_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/alexander_v__sc_conference_of_naacp_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/amgen_inc__v__sanofi_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/andy_warhol_found___inc__v__goldsmith_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/arizona_v__navajo_nation_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/becerra__sec__of_h_hs_v__san_carlos_apache_tribe_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/biden_v__nebraska_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/bissonnette_v__lepage_bakeries_park_st___llc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/bittner_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/brown_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/cantero_v__bank_of_america__n_a__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/cfpb_v__com__fin__services_assn__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/chiaverini_v__city_of_napoleon_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/ciminelli_v__united_state_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/city_of_grants_pass_v__johnson_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/coinbase__inc__v__bielski_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/coinbase__inc__v__suski_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/connelly_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/corner_post__inc__v__bd__of_governors__frs_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/counterman_v__colorado_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/cruz_v__arizona_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/culley_v__marshall_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/dept__of_agric__rural_dev__v__kirtz_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/dept__of_education_v__brown_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/dept__of_state_v__munoz_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/devillier_v__texas_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/diaz_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/dubin_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/dupree_v__younger_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/erlinger_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/fbi_v__fikre_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/fda_v__alliance_hippocratic_medicine_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/financial_oversight_board_v__cpi_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/fischer_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/garland__att_y_gen__v__cargill_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/glacier_northwest__inc__v__int_l_brotherhood_of_teamsters_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/gonzalez_v__google_llc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/gonzalez_v__trevino_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/great_lakes_insurance_se_v__raiders_retreat_realty_co___llc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/groff_v__dejoy_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/harrington_v__purdue_pharma_l_p__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/harrow_v__dept__of_defense_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/health_and_hospital_corp__v__talevski_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/helix_energy_solutions_v__hewitt_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/in_re_grand_jury_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/jack_daniel_s_properties__inc__v__vip_products_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/jones_v__hendrix_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/karcho_polselli_v__irs_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/lac_du_flambeau_band_v__coughlin_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/lindke_v__freed_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/loper_bright_enterprises__inc__v__raimondo__sec__of_comm__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/lora_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/macquarie_infrastructure_corp__v__moab_partners__l_p__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/mallory_v__norfolk_southern_railway_co__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/mcintosh_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/merrill_v__milligan_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/moore_v__harper_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/moore_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/moyle_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/muldrow_v__st__louis_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/murray_v__ubs_securities__llc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/murthy__surgeon_gen__v__missouri_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/netchoice__llc_v__paxton_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/new_york_v__new_jersey_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/nra_v__vullo_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/o_connor_ratcliff_v__garnier_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/oh_adjutant_gen__s_dept__v__flra_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/ohio_v__epa_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/perez_v__sturgis_public_schools_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/pugin_v__garland_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/pulsifer_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/relentless__inc__v__dept__of_commerce_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/rudisill_v__mcdonough__sec__of_va_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/sackett_v__epa_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/samia_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/santos_zacaria_v__garland__att_y_gen__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/sec_v__cochran_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/sec_v__jarkesy_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/sheetz_v__county_of_el_dorado_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/slack_technologies__llc_v__pirani_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/smith_v__arizona_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/smith_v__spizzirri_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/smith_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/snyder_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/starbucks_corp__v__mckinney_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/students_for_fair_admissions_v__university_of_nc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/texas_v__new_mexico_and_colorado_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/thornell_v__jones_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/truck_insurance_exchange_v__kaiser_gypsum_co__inc__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/trump_v__anderson_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/turkiye_halk_bankasi_a_s__v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/twitter__inc__v__taamneh_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/tyler_v__hennepin_county_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/u_s___ex_rel__polansky_v__executive_health_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/u_s___ex_rel__schutte_v__supervalu_inc__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/united_states_trustee_v__john_q__hammons_fall_2006__llc_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/united_states_v__hansen_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/united_states_v__rahimi_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/united_states_v__texas_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/vidal__under_sec__of_comm__v__elster_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/warner_chappell_music__inc__v__nealy_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/wilkins_v__united_states_sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/wilkinson_v__garland__att_y_gen__sharegpt.json
  - path: json
    type: sharegpt
    conversation: chatml
    data_files: train/argument_dataset/yegiazaryan_v__smagin_sharegpt.json

chat_template: chatml

unfrozen_parameters:
- ^lm_head.weight$
- ^model.embed_tokens.weight$
# input_layernorm layers
- model.layers.0.input_layernorm
- model.layers.1.input_layernorm
- model.layers.2.input_layernorm
- model.layers.3.input_layernorm
- model.layers.4.input_layernorm
- model.layers.5.input_layernorm
- model.layers.6.input_layernorm
- model.layers.7.input_layernorm
- model.layers.8.input_layernorm
- model.layers.9.input_layernorm
- model.layers.10.input_layernorm
- model.layers.11.input_layernorm
- model.layers.12.input_layernorm
- model.layers.13.input_layernorm
# mlp.down_proj layers
- model.layers.0.mlp.down_proj
- model.layers.1.mlp.down_proj
- model.layers.17.mlp.down_proj
- model.layers.19.mlp.down_proj
- model.layers.18.mlp.down_proj
- model.layers.5.mlp.down_proj
- model.layers.20.mlp.down_proj
- model.layers.2.mlp.down_proj
- model.layers.4.mlp.down_proj
- model.layers.6.mlp.down_proj
- model.layers.3.mlp.down_proj
- model.layers.16.mlp.down_proj
- model.layers.15.mlp.down_proj
- model.layers.13.mlp.down_proj
# mlp.gate_proj layers
- model.layers.0.mlp.gate_proj
- model.layers.1.mlp.gate_proj
- model.layers.2.mlp.gate_proj
- model.layers.3.mlp.gate_proj
- model.layers.22.mlp.gate_proj
- model.layers.21.mlp.gate_proj
- model.layers.20.mlp.gate_proj
- model.layers.23.mlp.gate_proj
- model.layers.19.mlp.gate_proj
- model.layers.4.mlp.gate_proj
- model.layers.18.mlp.gate_proj
- model.layers.17.mlp.gate_proj
- model.layers.5.mlp.gate_proj
- model.layers.24.mlp.gate_proj
# mlp.up_proj layers
- model.layers.4.mlp.up_proj
- model.layers.3.mlp.up_proj
- model.layers.5.mlp.up_proj
- model.layers.6.mlp.up_proj
- model.layers.7.mlp.up_proj
- model.layers.2.mlp.up_proj
- model.layers.8.mlp.up_proj
- model.layers.14.mlp.up_proj
- model.layers.13.mlp.up_proj
- model.layers.11.mlp.up_proj
- model.layers.9.mlp.up_proj
- model.layers.1.mlp.up_proj
- model.layers.15.mlp.up_proj
- model.layers.12.mlp.up_proj
# post_attention_layernorm layers
- model.layers.0.post_attention_layernorm
- model.layers.1.post_attention_layernorm
- model.layers.2.post_attention_layernorm
- model.layers.3.post_attention_layernorm
- model.layers.4.post_attention_layernorm
- model.layers.5.post_attention_layernorm
- model.layers.6.post_attention_layernorm
- model.layers.7.post_attention_layernorm
- model.layers.8.post_attention_layernorm
- model.layers.9.post_attention_layernorm
- model.layers.10.post_attention_layernorm
- model.layers.11.post_attention_layernorm
- model.layers.12.post_attention_layernorm
- model.layers.13.post_attention_layernorm
# self_attn.k_proj layers
- model.layers.25.self_attn.k_proj
- model.layers.22.self_attn.k_proj
- model.layers.19.self_attn.k_proj
- model.layers.20.self_attn.k_proj
- model.layers.17.self_attn.k_proj
- model.layers.24.self_attn.k_proj
- model.layers.23.self_attn.k_proj
- model.layers.18.self_attn.k_proj
- model.layers.21.self_attn.k_proj
- model.layers.27.self_attn.k_proj
- model.layers.15.self_attn.k_proj
- model.layers.10.self_attn.k_proj
- model.layers.6.self_attn.k_proj
- model.layers.5.self_attn.k_proj
# self_attn.o_proj layers
- model.layers.13.self_attn.o_proj
- model.layers.7.self_attn.o_proj
- model.layers.12.self_attn.o_proj
- model.layers.10.self_attn.o_proj
- model.layers.5.self_attn.o_proj
- model.layers.21.self_attn.o_proj
- model.layers.6.self_attn.o_proj
- model.layers.19.self_attn.o_proj
- model.layers.8.self_attn.o_proj
- model.layers.20.self_attn.o_proj
- model.layers.22.self_attn.o_proj
- model.layers.9.self_attn.o_proj
- model.layers.17.self_attn.o_proj
- model.layers.11.self_attn.o_proj
# self_attn.q_proj layers
- model.layers.12.self_attn.q_proj
- model.layers.13.self_attn.q_proj
- model.layers.9.self_attn.q_proj
- model.layers.8.self_attn.q_proj
- model.layers.10.self_attn.q_proj
- model.layers.14.self_attn.q_proj
- model.layers.11.self_attn.q_proj
- model.layers.15.self_attn.q_proj
- model.layers.26.self_attn.q_proj
- model.layers.6.self_attn.q_proj
- model.layers.7.self_attn.q_proj
- model.layers.16.self_attn.q_proj
- model.layers.5.self_attn.q_proj
- model.layers.25.self_attn.q_proj
# model.norm layers
# self_attn.v_proj layers
- model.layers.23.self_attn.v_proj
- model.layers.14.self_attn.v_proj
- model.layers.15.self_attn.v_proj
- model.layers.19.self_attn.v_proj
- model.layers.3.self_attn.v_proj
- model.layers.18.self_attn.v_proj
- model.layers.25.self_attn.v_proj
- model.layers.4.self_attn.v_proj
- model.layers.17.self_attn.v_proj
- model.layers.22.self_attn.v_proj
- model.layers.20.self_attn.v_proj
- model.layers.13.self_attn.v_proj
- model.layers.6.self_attn.v_proj
- model.layers.27.self_attn.v_proj

val_set_size: 0.05
output_dir: ./outputs/magistrate-3.2-3b

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

adapter: 

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 2e-4

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 1000
evals_per_epoch: 2
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed: deepspeed_configs/zero3.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|end_of_text|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

Model description

Magistrate-3.2-3b-it is a legal assistant specializing in US Supreme Court case law and US Federal regulations.

The base model is pretrained with ~250M tokens containing no synthetic legal data. The instruct model does contain synthetic data.

Intended uses & limitations

This model is for research purposes and for continued development of the legal specialty. You are liable for all model outputs.

Training and evaluation data

This model was trained on a variety of standard open source datasets like OpenHermes-2.5, hermes-function-calling, and some select entries from the Tome. Additionally, I have included a comprehensive, non-synthetic argument dataset. This is a work in progress but has shown promising results so far.

Training procedure

Spectrum top 35% finetune for both pretrain and SFT. Thanks to the cognitive computations team for the work done with spectrum.

Pretraining methodology based on Cohere's paper: To Code, or Not To Code? Exploring Impact of Code in Pre-training
Instruct finetune largely based on OpenHermes-2.5 and hermes-function-calling

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 8
total_train_batch_size: 16
total_eval_batch_size: 2
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 1000
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.3754	0.0005	1	1.7429
1.0	0.5002	1017	0.8864
0.9482	1.0005	2034	0.8395
0.6817	1.4987	3051	0.8063
0.697	1.9991	4068	0.7580
0.3769	2.4966	5085	0.8140
0.4278	2.9965	6102	0.8067

Framework versions

Transformers 4.45.0
Pytorch 2.3.1+cu121
Datasets 2.21.0
Tokenizers 0.20.0

Downloads last month: 87

GGUF

Model size

3B params

Architecture

llama

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Model tree for macadeliccc/magistrate-3.2-3b-it-GGUF

Base model

meta-llama/Llama-3.2-3B

Finetuned

macadeliccc/magistrate-3.2-3b-base

Quantized

(2)

this model

Datasets used to train macadeliccc/magistrate-3.2-3b-it-GGUF

Paper for macadeliccc/magistrate-3.2-3b-it-GGUF

To Code, or Not To Code? Exploring Impact of Code in Pre-training

Paper • 2408.10914 • Published Aug 20, 2024 • 45