Instructions to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer 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 jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S
# Run inference directly in the terminal:
llama cli -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S
# Run inference directly in the terminal:
llama cli -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S
# Run inference directly in the terminal:
./build/bin/llama cli -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Use Docker

docker model run hf.co/jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Llama new
LM Studio
Jan

vLLM

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Ollama
How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with Ollama:
```
ollama run hf.co/jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S
```
Unsloth Desktop

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with Docker Model Runner:
```
docker model run hf.co/jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S
```

Lemonade

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Run and chat with the model

lemonade run user.Swift-1.5-Flash-Next-Heretic-ReverseEngineer-IQ3_S

List all available models

lemonade list

Hermes Agent

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

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 jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S

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 "jackjusko/Swift-1.5-Flash-Next-Heretic-ReverseEngineer:IQ3_S" \
  --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"

looping

by BazsiBazsi - opened 3 days ago

Discussion

BazsiBazsi

3 days ago

I was kinda hyped for this model due to its size, but i got a lot of looping problems in real world tasks(with the IQ4XS-plus).
I was using strata, and i came from QFN Orca q4km. Any chance to requent to Q4KM? Maybe heavier sampling penalty?

jeffpeng3

3 days ago

I was wondering if you could provide some prompts that are likely to cause loops, so I can test them on other quantizations to see if the issue is caused by the model weights themselves.

WhiteZero

1 day ago

I was wondering if you could provide some prompts that are likely to cause loops, so I can test them on other quantizations to see if the issue is caused by the model weights themselves.

I can provide an example. Maybe it's a "gotcha" or otherwise not super relevant, but it's what I ran into today. I'm using the IQ3_S quant in Strata.
I downloaded this old 16-bit app and challenged this ReverseEngineer finetune to get it working for me: https://www.fourmilab.ch/moontoolw/moontool16.html
I prompted with "Inside the moontool project folder there is a Windows executable that won't run on this system. Diagnose why and see if we can use REA to make it compatible." inside of OpenCode running the latest public REA. Ultimately it kept getting stuck in this sort of loop below. I eventually interrupted it, switched to this model (https://huggingface.co/SC117/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF, same quant) and it didn't get stuck at all.


MS NE header: ne_cseg (segment count) is at 0x3A? The standard MS NE header fields and offsets (winnt.h IMAGE_NE_HEADER): WORD ne_magic; // 0 WORD ne_version; // 2 WORD ne_cc; // 4 WORD ne_heap; // 6 WORD ne_disk; // 8 WORD ne_apts; // 10 WORD ne_cbsel; // 12: offset of segment table WORD ne_idat?; // 14 WORD ne_flags?; // 16 WORD ne_cseg?; // 18 ... WORD ne_ventry?; // ...

MS NE header (winnt.h IMAGE_NE_HEADER): WORD ne_magic; WORD ne_version; WORD ne_cc; WORD ne_heap; WORD ne_disk; WORD ne_apts; WORD ne_cbsel; WORD ne_... WORD ne_flags? WORD ne_cseg?

winnt.h IMAGE_NE_HEADER: 0x00 WORD ne_magic 0x02 WORD ne_version 0x04 WORD ne_cc 0x06 WORD ne_heap 0x08 WORD ne_disk 0x0A WORD ne_apts 0x0C WORD ne_cbsel 0x0E WORD ne_idat 0x10 WORD ne_flags? 0x12 WORD ne_cseg? ... winnt.h IMAGE_NE_HEADER fields (MS): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags? WORD ne_cseg? WORD ne_...

MS NE header (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags? WORD ne_cseg?

winnt.h IMAGE_NE_HEADER: WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_initds? WORD ne_flags? WORD ne_cseg? WORD ne_...

MS NE header (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_idat WORD ne_flags? WORD ne_cseg? WORD ne_cdentry? WORD ne_...

winnt.h: typedef struct IMAGE_NE_HEADER { WORD ne_magic; WORD ne_version; WORD ne_cc; WORD ne_heap; WORD ne_disk; WORD ne_apts; WORD ne_cbsel; WORD ne_initds? WORD ne_flags? WORD ne_cseg? WORD ne... DWORD ne_... WORD ne_... WORD ne_... WORD ne_expoff? ... WORD ne_... WORD ne_res? ... }

The MS NE header is 62 bytes: 0x3E.

winnt.h IMAGE_NE_HEADER (MS): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags? WORD ne_cseg? DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_...

MS NE header fields (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_init? WORD ne_flags WORD ne_cseg WORD ne_... WORD ne_cdentry? WORD ne_... WORD ne_... WORD ne_expoff? WORD ne_... WORD ne_res? WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_heap? WORD ne_disk? WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_...

The MS NE header at 0x38 is: WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_idat WORD ne_flags? WORD ne_cseg? WORD ne_... WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_...

Let me map the values decoded from the dump (offset=decimal): 0x400 = 17742 (0x454E magic) 0x402 = 3589 (0x0E05 version 5.14) 0x404 = 427 (0x1AB) 0x406 = 40 (0x28) 0x408 = 5659 (0x161B) 0x40A = 10051 (0x2743) 0x40C = 770 (0x302) 0x40E = 2 0x410 = 4096 (0x1000) 0x412 = 5120 (0x1400) 0x414 = 17178 (0x431A) 0x416 = 1 0x418 = 0 0x41A = 2 0x41C = 2 0x41E = 4 0x420 = 56 (0x38) 0x422 = 64 (0x40) 0x424 = 80 (0x50) 0x426 = 364 (0x16C) 0x428 = 394 (0x18A) 0x42A = 402 (0x192) 0x42C = 1491 (0x5D3) 0x42E = 0 0x430 = 6 0x432 = 4 0x434 = 0 0x436 = 2050 (0x802) 0x438 = 98 (0x62) 0x43A = 3652 (0xE44) 0x43C = 0 0x43E = 768 (0x300) 0x440 = 100 (0x64) 0x442 = 35894 (0x8C36) 0x444 = 3408 (0x0D50) 0x446 = 35894 (0x8C36) 0x448 = 3312 (0x0CF0) 0x44A = 6928 (0x1B10) 0x44C = 3409 (0x0D51) 0x44E = 6928 (0x1B10) 0x450 = 4 0x452 = 32783 (0x800F) 0x454 = 1 0x456 = 0 0x458 = 0 0x45A = 3750 (0xEA6) 0x45C = 7 0x45E = 7216 (0x1C30) 0x460 = 32769 (0x8001) 0x462 = 0 0x464 = 0 0x466 = 32782 (0x800E) 0x468 = 1 0x46A = 0 0x46C = 0 0x46E = 3757 (0xEBD) 0x470 = 2 0x472 = 7216 (0x1C30) 0x474 = 32769 (0x8001) 0x476 = 0 0x478 = 0 0x47A = 32770 (0x8002) 0x47C = 2 0x47E = 0

MS NE header (winnt.h IMAGE_NE_HEADER) with MS field names: WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_idat WORD ne_flags WORD ne_cseg WORD ne_...

winnt.h IMAGE_NE_HEADER: WORD ne_magic; WORD ne_version; WORD ne_cc; WORD ne_heap; WORD ne_disk; WORD ne_apts; WORD ne_cbsel; WORD ne_init? WORD ne_flags? WORD ne_cseg? WORD ne_... WORD ne_... WORD ne_... WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... DWORD ne_heapinit? WORD ne_... WORD ne_... WORD ne_... WORD ne_...

winnt.h IMAGE_NE_HEADER (from MS): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags? WORD ne_cseg? WORD ne_cdentry? WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_...

MS NE header (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags WORD ne_cseg WORD ne_...

winnt.h IMAGE_NE_HEADER fields (actual): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_init? WORD ne_flags? WORD ne_cseg? WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_expoff? WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_res? WORD ne_... WORD ne_... WORD ne_...

MS NE header fields (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne_cbsel WORD ne_... WORD ne_flags? WORD ne_cseg? WORD ne_... WORD ne_... WORD ne_... WORD ne_... DWORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_... WORD ne_...

The MS NE header fields (winnt.h IMAGE_NE_HEADER) are: WORD ne_magic; WORD ne_version; WORD ne_cc; WORD ne_heap; WORD ne_disk; WORD ne_apts; WORD ne_cbsel; WORD ne_? WORD ne_flags; WORD ne_cseg; WORD ne_? WORD ne_? DWORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? DWORD ne_heap? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_? WORD ne_?

MS NE header (winnt.h IMAGE_NE_HEADER): WORD ne_magic WORD ne_version WORD ne_cc WORD ne_heap WORD ne_disk WORD ne_apts WORD ne```

jackjusko

Owner 1 day ago

Frankly, the IQ3_S GSQ-RCO quant uncensored is probably a higher quality than the IQ3_S quant I have in here, I originally optimized this design around the IQ4_NL build, and I added the IQ4_XS and IQ3_S upon request. Bumping up the sampling repeat penalty may help.

jackjusko changed discussion status to closed 1 day ago

BazsiBazsi

about 20 hours ago

Hey there. excuse for the late reply here. I tried to reproduce a couple of times this issue but eventually the agent stopped itself, the repetition loop went up to 1k+ tokens but it only happened in thinking. For an upper bound in prod i've set up a 32k token limit to catch these in case it eve happens again.
btw any cache to have dflash2 integrated in your next releases?

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