Instructions to use Crownelius/Crow-9B-HERETIC-4.6 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Crownelius/Crow-9B-HERETIC-4.6 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Crownelius/Crow-9B-HERETIC-4.6: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 Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Crownelius/Crow-9B-HERETIC-4.6: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 Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Use Docker

docker model run hf.co/Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

LM Studio
Jan
Ollama
How to use Crownelius/Crow-9B-HERETIC-4.6 with Ollama:
```
ollama run hf.co/Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
```

Unsloth Studio

How to use Crownelius/Crow-9B-HERETIC-4.6 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 Crownelius/Crow-9B-HERETIC-4.6 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 Crownelius/Crow-9B-HERETIC-4.6 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Crownelius/Crow-9B-HERETIC-4.6 to start chatting

How to use Crownelius/Crow-9B-HERETIC-4.6 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

OpenClaw new

How to use Crownelius/Crow-9B-HERETIC-4.6 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use Crownelius/Crow-9B-HERETIC-4.6 with Docker Model Runner:
```
docker model run hf.co/Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M
```

Lemonade

How to use Crownelius/Crow-9B-HERETIC-4.6 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Run and chat with the model

lemonade run user.Crow-9B-HERETIC-4.6-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Crownelius/Crow-9B-HERETIC-4.6 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Crownelius/Crow-9B-HERETIC-4.6:Q4_K_M

Run Hermes

hermes

Atomic Chat

Benchmarks

#14

by selimaktas - opened Mar 19

Discussion

selimaktas

Mar 19

Hello, I benchmarked this model and base Qwen3.5-9B using the lm evaluation harness with the leaderboard task, and it seems to get a worse score across all benches.

Crownelius/Crow-9B-HERETIC:

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
leaderboard	1.0	none		acc	↑	0.5258	±	0.0046
		none		acc_norm	↑	0.5696	±	0.0052
		none		exact_match	↑	0.3187	±	0.0118
		none		inst_level_loose_acc	↑	0.5935	±	N/A
		none		inst_level_strict_acc	↑	0.5600	±	N/A
		none		prompt_level_loose_acc	↑	0.4399	±	0.0214
		none		prompt_level_strict_acc	↑	0.3974	±	0.0211
- leaderboard_bbh		none		acc_norm	↑	0.6178	±	0.0059
- leaderboard_bbh_boolean_expressions	1.0	none	3	acc_norm	↑	0.8440	±	0.0230
- leaderboard_bbh_causal_judgement	1.0	none	3	acc_norm	↑	0.5615	±	0.0364
- leaderboard_bbh_date_understanding	1.0	none	3	acc_norm	↑	0.7240	±	0.0283
- leaderboard_bbh_disambiguation_qa	1.0	none	3	acc_norm	↑	0.6040	±	0.0310
- leaderboard_bbh_formal_fallacies	1.0	none	3	acc_norm	↑	0.4640	±	0.0316
- leaderboard_bbh_geometric_shapes	1.0	none	3	acc_norm	↑	0.4960	±	0.0317
- leaderboard_bbh_hyperbaton	1.0	none	3	acc_norm	↑	0.6560	±	0.0301
- leaderboard_bbh_logical_deduction_five_objects	1.0	none	3	acc_norm	↑	0.5720	±	0.0314
- leaderboard_bbh_logical_deduction_seven_objects	1.0	none	3	acc_norm	↑	0.5320	±	0.0316
- leaderboard_bbh_logical_deduction_three_objects	1.0	none	3	acc_norm	↑	0.7720	±	0.0266
- leaderboard_bbh_movie_recommendation	1.0	none	3	acc_norm	↑	0.8600	±	0.0220
- leaderboard_bbh_navigate	1.0	none	3	acc_norm	↑	0.7320	±	0.0281
- leaderboard_bbh_object_counting	1.0	none	3	acc_norm	↑	0.4640	±	0.0316
- leaderboard_bbh_penguins_in_a_table	1.0	none	3	acc_norm	↑	0.6438	±	0.0398
- leaderboard_bbh_reasoning_about_colored_objects	1.0	none	3	acc_norm	↑	0.7720	±	0.0266
- leaderboard_bbh_ruin_names	1.0	none	3	acc_norm	↑	0.8360	±	0.0235
- leaderboard_bbh_salient_translation_error_detection	1.0	none	3	acc_norm	↑	0.6560	±	0.0301
- leaderboard_bbh_snarks	1.0	none	3	acc_norm	↑	0.6124	±	0.0366
- leaderboard_bbh_sports_understanding	1.0	none	3	acc_norm	↑	0.7880	±	0.0259
- leaderboard_bbh_temporal_sequences	1.0	none	3	acc_norm	↑	0.9200	±	0.0172
- leaderboard_bbh_tracking_shuffled_objects_five_objects	1.0	none	3	acc_norm	↑	0.2560	±	0.0277
- leaderboard_bbh_tracking_shuffled_objects_seven_objects	1.0	none	3	acc_norm	↑	0.1640	±	0.0235
- leaderboard_bbh_tracking_shuffled_objects_three_objects	1.0	none	3	acc_norm	↑	0.3400	±	0.0300
- leaderboard_bbh_web_of_lies	1.0	none	3	acc_norm	↑	0.5520	±	0.0315
- leaderboard_gpqa		none		acc_norm	↑	0.4060	±	0.0142
- leaderboard_gpqa_diamond	1.0	none	0	acc_norm	↑	0.4293	±	0.0353
- leaderboard_gpqa_extended	1.0	none	0	acc_norm	↑	0.4048	±	0.0210
- leaderboard_gpqa_main	1.0	none	0	acc_norm	↑	0.3973	±	0.0231
- leaderboard_ifeval	3.0	none	0	inst_level_loose_acc	↑	0.5935	±	N/A
		none	0	inst_level_strict_acc	↑	0.5600	±	N/A
		none	0	prompt_level_loose_acc	↑	0.4399	±	0.0214
		none	0	prompt_level_strict_acc	↑	0.3974	±	0.0211
- leaderboard_math_hard		none		exact_match	↑	0.3187	±	0.0118
- leaderboard_math_algebra_hard	3.0	none	4	exact_match	↑	0.5635	±	0.0284
		none	4	exact_match_original	↑	0.5570	±	0.0284
- leaderboard_math_counting_and_prob_hard	3.0	none	4	exact_match	↑	0.2602	±	0.0397
		none	4	exact_match_original	↑	0.2683	±	0.0401
- leaderboard_math_geometry_hard	3.0	none	4	exact_match	↑	0.1515	±	0.0313
		none	4	exact_match_original	↑	0.1439	±	0.0307
- leaderboard_math_intermediate_algebra_hard	3.0	none	4	exact_match	↑	0.1250	±	0.0198
		none	4	exact_match_original	↑	0.1286	±	0.0200
- leaderboard_math_num_theory_hard	3.0	none	4	exact_match	↑	0.3052	±	0.0372
		none	4	exact_match_original	↑	0.3052	±	0.0372
- leaderboard_math_prealgebra_hard	3.0	none	4	exact_match	↑	0.5130	±	0.0361
		none	4	exact_match_original	↑	0.5130	±	0.0361
- leaderboard_math_precalculus_hard	3.0	none	4	exact_match	↑	0.1185	±	0.0279
		none	4	exact_match_original	↑	0.1111	±	0.0271
- leaderboard_mmlu_pro	0.1	none	5	acc	↑	0.5258	±	0.0046
- leaderboard_musr		none		acc_norm	↑	0.4603	±	0.0180
- leaderboard_musr_murder_mysteries	1.0	none	0	acc_norm	↑	0.5480	±	0.0315
- leaderboard_musr_object_placements	1.0	none	0	acc_norm	↑	0.3867	±	0.0305
- leaderboard_musr_team_allocation	1.0	none	0	acc_norm	↑	0.4480	±	0.0315

Groups	Version	Filter	Metric		Value		Stderr
leaderboard	1	none	acc	↑	0.5258	±	0.0046
		none	acc_norm	↑	0.5696	±	0.0052
		none	exact_match	↑	0.3187	±	0.0118
		none	inst_level_loose_acc	↑	0.5935	±	N/A
		none	inst_level_strict_acc	↑	0.5600	±	N/A
		none	prompt_level_loose_acc	↑	0.4399	±	0.0214
		none	prompt_level_strict_acc	↑	0.3974	±	0.0211
- leaderboard_bbh		none	acc_norm	↑	0.6178	±	0.0059
- leaderboard_gpqa		none	acc_norm	↑	0.4060	±	0.0142
- leaderboard_math_hard		none	exact_match	↑	0.3187	±	0.0118
- leaderboard_musr		none	acc_norm	↑	0.4603	±	0.0180

Qwen/Qwen3.5-9B:

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
leaderboard	1.0	none		acc	↑	0.5490	±	0.0045
		none		acc_norm	↑	0.5739	±	0.0052
		none		exact_match	↑	0.3965	±	0.0128
		none		inst_level_loose_acc	↑	0.6379	±	N/A
		none		inst_level_strict_acc	↑	0.6163	±	N/A
		none		prompt_level_loose_acc	↑	0.5083	±	0.0215
		none		prompt_level_strict_acc	↑	0.4806	±	0.0215
- leaderboard_bbh		none		acc_norm	↑	0.6190	±	0.0058
- leaderboard_bbh_boolean_expressions	1.0	none	3	acc_norm	↑	0.8400	±	0.0232
- leaderboard_bbh_causal_judgement	1.0	none	3	acc_norm	↑	0.5775	±	0.0362
- leaderboard_bbh_date_understanding	1.0	none	3	acc_norm	↑	0.7680	±	0.0268
- leaderboard_bbh_disambiguation_qa	1.0	none	3	acc_norm	↑	0.6200	±	0.0308
- leaderboard_bbh_formal_fallacies	1.0	none	3	acc_norm	↑	0.7200	±	0.0285
- leaderboard_bbh_geometric_shapes	1.0	none	3	acc_norm	↑	0.4840	±	0.0317
- leaderboard_bbh_hyperbaton	1.0	none	3	acc_norm	↑	0.6400	±	0.0304
- leaderboard_bbh_logical_deduction_five_objects	1.0	none	3	acc_norm	↑	0.5800	±	0.0313
- leaderboard_bbh_logical_deduction_seven_objects	1.0	none	3	acc_norm	↑	0.5320	±	0.0316
- leaderboard_bbh_logical_deduction_three_objects	1.0	none	3	acc_norm	↑	0.7680	±	0.0268
- leaderboard_bbh_movie_recommendation	1.0	none	3	acc_norm	↑	0.7960	±	0.0255
- leaderboard_bbh_navigate	1.0	none	3	acc_norm	↑	0.6480	±	0.0303
- leaderboard_bbh_object_counting	1.0	none	3	acc_norm	↑	0.4240	±	0.0313
- leaderboard_bbh_penguins_in_a_table	1.0	none	3	acc_norm	↑	0.6712	±	0.0390
- leaderboard_bbh_reasoning_about_colored_objects	1.0	none	3	acc_norm	↑	0.7880	±	0.0259
- leaderboard_bbh_ruin_names	1.0	none	3	acc_norm	↑	0.8560	±	0.0222
- leaderboard_bbh_salient_translation_error_detection	1.0	none	3	acc_norm	↑	0.6760	±	0.0297
- leaderboard_bbh_snarks	1.0	none	3	acc_norm	↑	0.6236	±	0.0364
- leaderboard_bbh_sports_understanding	1.0	none	3	acc_norm	↑	0.8040	±	0.0252
- leaderboard_bbh_temporal_sequences	1.0	none	3	acc_norm	↑	0.9240	±	0.0168
- leaderboard_bbh_tracking_shuffled_objects_five_objects	1.0	none	3	acc_norm	↑	0.1880	±	0.0248
- leaderboard_bbh_tracking_shuffled_objects_seven_objects	1.0	none	3	acc_norm	↑	0.1520	±	0.0228
- leaderboard_bbh_tracking_shuffled_objects_three_objects	1.0	none	3	acc_norm	↑	0.3000	±	0.0290
- leaderboard_bbh_web_of_lies	1.0	none	3	acc_norm	↑	0.4880	±	0.0317
- leaderboard_gpqa		none		acc_norm	↑	0.4446	±	0.0144
- leaderboard_gpqa_diamond	1.0	none	0	acc_norm	↑	0.4394	±	0.0354
- leaderboard_gpqa_extended	1.0	none	0	acc_norm	↑	0.4524	±	0.0213
- leaderboard_gpqa_main	1.0	none	0	acc_norm	↑	0.4375	±	0.0235
- leaderboard_ifeval	3.0	none	0	inst_level_loose_acc	↑	0.6379	±	N/A
		none	0	inst_level_strict_acc	↑	0.6163	±	N/A
		none	0	prompt_level_loose_acc	↑	0.5083	±	0.0215
		none	0	prompt_level_strict_acc	↑	0.4806	±	0.0215
- leaderboard_math_hard		none		exact_match	↑	0.3965	±	0.0128
- leaderboard_math_algebra_hard	3.0	none	4	exact_match	↑	0.5798	±	0.0282
		none	4	exact_match_original	↑	0.2638	±	0.0252
- leaderboard_math_counting_and_prob_hard	3.0	none	4	exact_match	↑	0.3496	±	0.0432
		none	4	exact_match_original	↑	0.1220	±	0.0296
- leaderboard_math_geometry_hard	3.0	none	4	exact_match	↑	0.3182	±	0.0407
		none	4	exact_match_original	↑	0.2045	±	0.0352
- leaderboard_math_intermediate_algebra_hard	3.0	none	4	exact_match	↑	0.1786	±	0.0229
		none	4	exact_match_original	↑	0.1286	±	0.0200
- leaderboard_math_num_theory_hard	3.0	none	4	exact_match	↑	0.4740	±	0.0404
		none	4	exact_match_original	↑	0.2078	±	0.0328
- leaderboard_math_prealgebra_hard	3.0	none	4	exact_match	↑	0.5078	±	0.0361
		none	4	exact_match_original	↑	0.1762	±	0.0275
- leaderboard_math_precalculus_hard	3.0	none	4	exact_match	↑	0.3037	±	0.0397
		none	4	exact_match_original	↑	0.2296	±	0.0363
- leaderboard_mmlu_pro	0.1	none	5	acc	↑	0.5490	±	0.0045
- leaderboard_musr		none		acc_norm	↑	0.4339	±	0.0176
- leaderboard_musr_murder_mysteries	1.0	none	0	acc_norm	↑	0.5440	±	0.0316
- leaderboard_musr_object_placements	1.0	none	0	acc_norm	↑	0.2891	±	0.0284
- leaderboard_musr_team_allocation	1.0	none	0	acc_norm	↑	0.4720	±	0.0316

Groups	Version	Filter	Metric		Value		Stderr
leaderboard	1	none	acc	↑	0.5490	±	0.0045
		none	acc_norm	↑	0.5739	±	0.0052
		none	exact_match	↑	0.3965	±	0.0128
		none	inst_level_loose_acc	↑	0.6379	±	N/A
		none	inst_level_strict_acc	↑	0.6163	±	N/A
		none	prompt_level_loose_acc	↑	0.5083	±	0.0215
		none	prompt_level_strict_acc	↑	0.4806	±	0.0215
- leaderboard_bbh		none	acc_norm	↑	0.6190	±	0.0058
- leaderboard_gpqa		none	acc_norm	↑	0.4446	±	0.0144
- leaderboard_math_hard		none	exact_match	↑	0.3965	±	0.0128
- leaderboard_musr		none	acc_norm	↑	0.4339	±	0.0176

selimaktas

Mar 19

Benchmarked with vLLM using the following command:
lm_eval
--model local-completions
--model_args "base_url=http://localhost:8000/v1/completions,pretrained=Crownelius/Crow-9B-HERETIC,tensor_parallel_size=1,add_bos_token=true,trust_remote_code=true,max_length=24576,max_gen_toks=16384,tokenizer=Qwen/Qwen3.5-9B,enable_thinking=True"
--tasks leaderboard
--batch_size 1024
--seed 42

geekmemz

Mar 20

That`s right.I think it is worst than origin model when i using this crow 9b to code.Certainly,it is not a best practice to use such datasets including writing,coding etc.
A specific dataset can boost a part of ability example coding,a too diversity dataset will get worse.

selimaktas

Mar 20

That`s right.I think it is worst than origin model when i using this crow 9b to code.Certainly,it is not a best practice to use such datasets including writing,coding etc.
A specific dataset can boost a part of ability example coding,a too diversity dataset will get worse.

None of these benchmarks are code-related, and I wouldn't say the degradation is too much. These are just general tests to compare the model to the original.

Crownelius

Owner Mar 20

Honestly this is normal with training with two different reasoning types. As long as it's not a huge drop, it's negligible

selimaktas

Mar 21

I mostly agree except the math task.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment