Instructions to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B")
model = PeftModel.from_pretrained(base_model, "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b")

Transformers

How to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b with Transformers:

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

pipe = pipeline("text-generation", model="adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b

SGLang

How to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b 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 "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b with Docker Model Runner:
```
docker model run hf.co/adamkarvonen/checkpoints_cont_50k_hbi_50k_mix_qwen3_14b
```

checkpoints_cont_50k_hbi_50k_mix_qwen3_14b / ao_config.json

adamkarvonen

Upload folder using huggingface_hub

e78ab7c verified 5 months ago

Raw

History Blame Contribute Delete

15.5 kB

	{
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"act_layer_combinations": [
	[
	10,
	20,
	30
	]
	],
	"schema_version": 1,
	"special_token": " ?",
	"prefix_template": "Layer: {layer}\\n{special_token} * {num_positions} \\n",
	"model_name": "Qwen/Qwen3-14B",
	"hook_onto_layer": 1,
	"dataset_configs": [
	{
	"custom_dataset_params": {
	"min_k_tokens": 1,
	"max_k_tokens": 50,
	"min_k_activations": 1,
	"max_k_activations": 1,
	"max_length": 2000,
	"directions": [
	"past",
	"future"
	],
	"vllm_max_new_tokens": 200,
	"max_vllm_context_tokens": 2000,
	"future_chat_system_prompt_prob": 0.5,
	"system_prompt_path": "datasets/latentqa_datasets/train/system.json",
	"english_only_temp_filter": false
	},
	"num_train": 5000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "past_lens",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"min_k_tokens": 1,
	"max_k_tokens": 50,
	"min_k_activations": 1,
	"max_k_activations": 50,
	"max_length": 2000,
	"directions": [
	"past",
	"future"
	],
	"vllm_max_new_tokens": 200,
	"max_vllm_context_tokens": 2000,
	"future_chat_system_prompt_prob": 0.5,
	"system_prompt_path": "datasets/latentqa_datasets/train/system.json",
	"english_only_temp_filter": false
	},
	"num_train": 5000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "past_lens",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"data_path": "datasets/latentqa_datasets/artifacts/Qwen3-8B/spqa_v2.json"
	},
	"num_train": 5000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "synthetic_qa",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "geometry_of_truth",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_geometry_of_truth",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "geometry_of_truth",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_geometry_of_truth",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "relations",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_relations",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "relations",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_relations",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "sst2",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_sst2",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "sst2",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_sst2",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "md_gender",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_md_gender",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "md_gender",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_md_gender",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "snli",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_snli",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "snli",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_snli",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "ner",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_ner",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "ner",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_ner",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "tense",
	"num_qa_per_sample": 2,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 1,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_tense",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"classification_dataset_name": "tense",
	"num_qa_per_sample": 1,
	"min_end_offset": -1,
	"max_end_offset": -5,
	"max_window_size": 50,
	"min_window_size": 1
	},
	"num_train": 1000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "classification_tense",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"data_path": "datasets/training_data/artifacts/synthetic_qa_2gpu_100k_qwen3_14b/training_data.json"
	},
	"num_train": 10000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "synthetic_qa",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"data_path": "datasets/training_data/artifacts/system_prompt_qa/hidden_bias_v2_qwen3_14b/training_data.json"
	},
	"num_train": 10000,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "synthetic_qa",
	"dataset_folder": "sft_training_data",
	"seed": 42
	},
	{
	"custom_dataset_params": {
	"data_path": "datasets/training_data/artifacts/system_prompt_qa/hidden_bias_inactive_qwen3_14b/training_data.json"
	},
	"num_train": 49950,
	"num_test": 0,
	"splits": [
	"train"
	],
	"model_name": "Qwen/Qwen3-14B",
	"layer_combinations": [
	[
	25,
	50,
	75
	]
	],
	"save_acts": false,
	"batch_size": 16,
	"dataset_name": "synthetic_qa",
	"dataset_folder": "sft_training_data",
	"seed": 42
	}
	],
	"dataset_loader_names": [
	"past_lens",
	"past_lens",
	"synthetic_qa",
	"classification_geometry_of_truth",
	"classification_geometry_of_truth",
	"classification_relations",
	"classification_relations",
	"classification_sst2",
	"classification_sst2",
	"classification_md_gender",
	"classification_md_gender",
	"classification_snli",
	"classification_snli",
	"classification_ner",
	"classification_ner",
	"classification_tense",
	"classification_tense",
	"synthetic_qa",
	"synthetic_qa",
	"synthetic_qa"
	],
	"use_decoder_vectors": true,
	"generation_kwargs": {
	"do_sample": false,
	"max_new_tokens": 20
	},
	"steering_coefficient": 1.0,
	"dataset_folder": "sft_training_data",
	"chat_regularization_path": null,
	"chat_regularization_every_n_ao_updates": 0,
	"chat_regularization_weight": 1.0,
	"chat_regularization_max_train_examples": null,
	"train_batch_size": 16,
	"eval_batch_size": 128,
	"train_batches_per_materialization_block": 64,
	"use_lora": true,
	"lora_r": 64,
	"lora_alpha": 128,
	"lora_dropout": 0.05,
	"lora_target_modules": "all-linear",
	"num_epochs": 1,
	"lr": 5e-06,
	"gradient_accumulation_steps": 1,
	"max_grad_norm": 1.0,
	"eval_steps": 2000,
	"eval_on_start": false,
	"gradient_checkpointing": true,
	"window_mult": 20,
	"save_steps": 2000,
	"save_dir": "checkpoints/cont_50k_hbi_50k_mix_qwen3_14b",
	"max_train_examples": null,
	"seed": 42,
	"eval_logs_path": "eval_logs.json",
	"load_lora_path": "checkpoints/500k_pl_31k_spqav2_199k_sqav3_50k_hb_126k_cls_qwen3_14b/final",
	"created_at_utc": "",
	"git_commit": "",
	"wandb_project": "misc_experiments",
	"wandb_run_name": "cont_50k_hbi_50k_mix_qwen3_14b",
	"wandb_suffix": "_cont_50k_hbi_50k_mix_qwen3_14b",
	"hf_push_to_hub": false,
	"hf_private_repo": false,
	"hf_repo_name": "",
	"hf_repo_id": "",
	"load_in_8bit": false,
	"open_ended_eval_include": [
	"number_prediction",
	"mmlu_prediction",
	"backtracking",
	"missing_info",
	"sycophancy",
	"system_prompt_qa_hidden",
	"system_prompt_qa_latentqa"
	],
	"positive_negative_examples": false
	}