Instructions to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1",
	filename="ornith-1.0-35b-IQ4_XS-MTP-graft-headQ6.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

Use Docker

docker model run hf.co/LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

LM Studio
Jan

vLLM

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

Ollama
How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with Ollama:
```
ollama run hf.co/LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
```

Unsloth Studio

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 to start chatting

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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": "LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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 "LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1: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 LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with Docker Model Runner:
```
docker model run hf.co/LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M
```

Lemonade

How to use LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull LordNeel/Ornith-1.0-35B-GGUF-llamacpp-tp1:Q4_K_M

Run and chat with the model

lemonade run user.Ornith-1.0-35B-GGUF-llamacpp-tp1-Q4_K_M

List all available models

lemonade list

LordNeel commited on 25 days ago

Commit

033d6f3

verified ·

1 Parent(s): b42a677

Update MTP benchmark profile and serving card

Browse files

Files changed (1) hide show

README.md +11 -6

README.md CHANGED Viewed

@@ -62,7 +62,7 @@ Use `Q3_K_M` when footprint matters more than peak throughput. It is a custom qu
 Use `IQ4_XS` when you want a middle-footprint quant: smaller than Q4_K_M with better next-token KLD than Q3_K_M, but slower than Q4_K_M in the short-context throughput profile.
-Use the trained Q6_K MTP draft only for MTP experimentation and regression work. It makes llama.cpp `draft-mtp` operational with 67.22% draft-token acceptance at `n_max=2`, but it is not yet a production throughput speedup: the measured IQ4_XS target plus Q6_K draft profile reached 286.77 tok/s at c16 versus 482.68 tok/s for refreshed IQ4_XS target-only. The smaller Q5_K_M and IQ4_XS MTP drafts are included as experimental footprint variants but are not recommended.
 ## Quant Provenance
@@ -137,6 +137,7 @@ Experimental trained-MTP profile:
 ```bash
 SPEC_DRAFT_BACKEND_SAMPLING=1 \
 LLAMA_MTP_FAST_BACKEND_SAMPLE=1 \
 LLAMA_MTP_DRAFT_TOP_K=1 LLAMA_MTP_DRAFT_TOP_P=1 LLAMA_MTP_DRAFT_TEMP=1 \
 QUANT=IQ4_XS PORT=8002 CTX_SIZE=8192 PARALLEL=16 \
@@ -223,12 +224,13 @@ OpenAI-compatible correctness gate, but only accepted 1.21% of draft tokens.
 This lab then trained a chain-corrected Qwen3.5 MoE MTP draft from target
 hidden-state captures and exported Q6_K, Q5_K_M, and IQ4_XS draft GGUFs. The
 current best MTP profile is IQ4_XS target plus Q6_K trained draft with
-`n_max=2`, draft backend sampling, and fast top-k=1 draft sampling:
-- c16 throughput: 286.77 tok/s
-- draft-token acceptance: 67.22%
-- acceptance by position: `(0.719, 0.621)`
-- approximate loaded VRAM: 22.44 GiB
 This is operational but still not a production MTP speedup. The refreshed
 IQ4_XS target-only profile reaches 482.68 tok/s at c16. The remaining blocker is
@@ -289,10 +291,13 @@ Experimental MTP profiles on the same IQ4_XS target:
 | IQ4_XS + BF16 MTP draft, n_max=3 | 25.12 | 95.5 | 97.0 | 102.1 | 108.2 | 0.47% |
 | IQ4_XS + BF16 MTP draft, n_max=1 | 23.03 | 153.6 | 147.2 | 154.3 | 159.5 | 1.07% |
 | IQ4_XS + IQ4_XS MTP draft, n_max=1 | 21.32 | 155.1 | 151.6 | 155.8 | 162.7 | 1.21% |
 | IQ4_XS + Q6_K trained MTP draft, n_max=2 | 22.44 | n/t | n/t | n/t | 286.77 | 67.22% |
 | IQ4_XS + Q5_K_M trained MTP draft, n_max=2 | 22.16 | n/t | n/t | n/t | 285.32 | 66.61% |
 | IQ4_XS + IQ4_XS trained MTP draft, n_max=2 | 21.53 | n/t | n/t | n/t | 265.06 | 59.20% |
 | IQ4_XS + Q6_K trained MTP draft, n_max=1 | 22.44 | n/t | n/t | n/t | 262.76 | 68.12% |
 `n/t` means not tested in the latest trained-MTP loop; those rows were measured
 only on the c16 profile used for the MTP fix loop.

 Use `IQ4_XS` when you want a middle-footprint quant: smaller than Q4_K_M with better next-token KLD than Q3_K_M, but slower than Q4_K_M in the short-context throughput profile.
+Use the trained Q6_K MTP draft only for MTP experimentation and regression work. It makes llama.cpp `draft-mtp` operational with 67.19-67.22% draft-token acceptance at `n_max=2`, but it is not yet a production throughput speedup: the fastest measured IQ4_XS target plus Q6_K draft profile reached 288.67 tok/s at c16 with per-head MTP contexts, versus 482.68 tok/s for refreshed IQ4_XS target-only. The smaller Q5_K_M and IQ4_XS MTP drafts are included as experimental footprint variants but are not recommended.
 ## Quant Provenance
 ```bash
 SPEC_DRAFT_BACKEND_SAMPLING=1 \
+MTP_PER_HEAD_CONTEXT=1 \
 LLAMA_MTP_FAST_BACKEND_SAMPLE=1 \
 LLAMA_MTP_DRAFT_TOP_K=1 LLAMA_MTP_DRAFT_TOP_P=1 LLAMA_MTP_DRAFT_TEMP=1 \
 QUANT=IQ4_XS PORT=8002 CTX_SIZE=8192 PARALLEL=16 \
 This lab then trained a chain-corrected Qwen3.5 MoE MTP draft from target
 hidden-state captures and exported Q6_K, Q5_K_M, and IQ4_XS draft GGUFs. The
 current best MTP profile is IQ4_XS target plus Q6_K trained draft with
+`n_max=2`, per-head MTP contexts, draft backend sampling, and fast top-k=1
+draft sampling:
+- c16 throughput: 288.67 tok/s
+- draft-token acceptance: 67.19%
+- acceptance by position: `(0.718, 0.621)`
+- approximate loaded VRAM: about 25.5 GiB
 This is operational but still not a production MTP speedup. The refreshed
 IQ4_XS target-only profile reaches 482.68 tok/s at c16. The remaining blocker is
 | IQ4_XS + BF16 MTP draft, n_max=3 | 25.12 | 95.5 | 97.0 | 102.1 | 108.2 | 0.47% |
 | IQ4_XS + BF16 MTP draft, n_max=1 | 23.03 | 153.6 | 147.2 | 154.3 | 159.5 | 1.07% |
 | IQ4_XS + IQ4_XS MTP draft, n_max=1 | 21.32 | 155.1 | 151.6 | 155.8 | 162.7 | 1.21% |
+| IQ4_XS + Q6_K trained MTP draft, n_max=2, per-head contexts | ~25.5 | n/t | n/t | n/t | 288.67 | 67.19% |
 | IQ4_XS + Q6_K trained MTP draft, n_max=2 | 22.44 | n/t | n/t | n/t | 286.77 | 67.22% |
 | IQ4_XS + Q5_K_M trained MTP draft, n_max=2 | 22.16 | n/t | n/t | n/t | 285.32 | 66.61% |
 | IQ4_XS + IQ4_XS trained MTP draft, n_max=2 | 21.53 | n/t | n/t | n/t | 265.06 | 59.20% |
 | IQ4_XS + Q6_K trained MTP draft, n_max=1 | 22.44 | n/t | n/t | n/t | 262.76 | 68.12% |
+| IQ4_XS + Q6_K trained MTP draft, n_max=2, batched deferred replay | 22.44 | n/t | n/t | n/t | 282.51 | 67.10% |
+| IQ4_XS + Q6_K trained MTP draft, n_max=3 | 22.44 | n/t | n/t | n/t | 270.19 | 58.54% |
 `n/t` means not tested in the latest trained-MTP loop; those rows were measured
 only on the c16 profile used for the MTP fix loop.