Instructions to use CMSManhattan/JiRackUltra_1b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use CMSManhattan/JiRackUltra_1b 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 CMSManhattan/JiRackUltra_1b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CMSManhattan/JiRackUltra_1b:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CMSManhattan/JiRackUltra_1b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CMSManhattan/JiRackUltra_1b: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 CMSManhattan/JiRackUltra_1b:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf CMSManhattan/JiRackUltra_1b: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 CMSManhattan/JiRackUltra_1b:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf CMSManhattan/JiRackUltra_1b:Q4_K_M

Use Docker

docker model run hf.co/CMSManhattan/JiRackUltra_1b:Q4_K_M

LM Studio
Jan

vLLM

How to use CMSManhattan/JiRackUltra_1b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CMSManhattan/JiRackUltra_1b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CMSManhattan/JiRackUltra_1b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CMSManhattan/JiRackUltra_1b:Q4_K_M

Ollama
How to use CMSManhattan/JiRackUltra_1b with Ollama:
```
ollama run hf.co/CMSManhattan/JiRackUltra_1b:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use CMSManhattan/JiRackUltra_1b with Docker Model Runner:
```
docker model run hf.co/CMSManhattan/JiRackUltra_1b:Q4_K_M
```

Lemonade

How to use CMSManhattan/JiRackUltra_1b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull CMSManhattan/JiRackUltra_1b:Q4_K_M

Run and chat with the model

lemonade run user.JiRackUltra_1b-Q4_K_M

List all available models

lemonade list

Atomic Chat

kgrabko commited on 10 days ago

Commit

c66a229

verified ·

1 Parent(s): 8f857e3

Initial upload of JiRackTernary Pro 10B weights

Browse files

Files changed (3) hide show

.gitattributes +1 -0
JiRackTernaryUltra_1b.py +9 -10
JiRackUltra_1b.Q6_K.gguf +3 -0

.gitattributes CHANGED Viewed

@@ -38,3 +38,4 @@ JiRackUltra_1b.gguf filter=lfs diff=lfs merge=lfs -text
 JiRackUltra_1b_Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 JiRackUltra_1b_Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text

 JiRackUltra_1b_Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 JiRackUltra_1b_Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+JiRackUltra_1b.Q6_K.gguf filter=lfs diff=lfs merge=lfs -text

JiRackTernaryUltra_1b.py CHANGED Viewed

@@ -1,19 +1,18 @@
-#%%writefile JiRackTernaryUltra_1b.py
 # =============================================================================
 # COPYRIGHT © 2026 Konstantin Vladimirovich Grabko. ALL RIGHTS RESERVED.
 # JiRack Ultra Ternary Transformer
 #
 # CMS Manhattan JiRack Technology — PATENT PENDING
 #
-# This code is proprietary.
 # Personal and non-commercial research use is allowed.
-# Any commercial use, derivative works for profit, or distribution
 # requires a paid license and 5% royalty.
 #
 # Unauthorized commercial use is strictly prohibited.
 # Contact: grabko@cmsmanhattan.com
 # =============================================================================
 import math
 import torch
@@ -365,12 +364,14 @@ class JiRackTransformer(nn.Module):
 # Convenience aliases so existing training scripts barely change:
-JiRackConfig = JiRackConfig          # drop-in name compat (optional)
 JiRackTransformer = JiRackTransformer
 # =============================================================================
-# Smoke test:  python JiRackTernaryUltra_1b.py   (tiny config, CPU, seconds)
 # =============================================================================
 if __name__ == "__main__":
     class TinyConfig(JiRackConfig):
@@ -420,9 +421,7 @@ if __name__ == "__main__":
     # 4) lambda survives a state_dict round-trip.
     sd = model.state_dict()
-    # [BUGFIX-1b] Was: JiRackTransformerDS1p5B(TinyConfig()) — a class name left
-    # over from the old ds1p5b file that does not exist here (NameError).
-    model2 = JiRackTransformer(TinyConfig())
     model2.load_state_dict(sd)
     assert abs(model2.get_lambda() - 1.0) < 1e-9, "lambda_ not serialized!"
     print("lambda serialization: OK")
@@ -467,4 +466,4 @@ if __name__ == "__main__":
     print(f"export: {sum('codes' in k for k in exported)} ternary tensors, "
           f"biases preserved, OK")
-    print("\nAll smoke tests passed.")

+#%%writefile JiRackTernaryUltra_1p5b.py
 # =============================================================================
 # COPYRIGHT © 2026 Konstantin Vladimirovich Grabko. ALL RIGHTS RESERVED.
 # JiRack Ultra Ternary Transformer
 #
 # CMS Manhattan JiRack Technology — PATENT PENDING
 #
+# This code is proprietary.
 # Personal and non-commercial research use is allowed.
+# Any commercial use, derivative works for profit, or distribution
 # requires a paid license and 5% royalty.
 #
 # Unauthorized commercial use is strictly prohibited.
 # Contact: grabko@cmsmanhattan.com
 # =============================================================================
 import math
 import torch
 # Convenience aliases so existing training scripts barely change:
+JiRackConfig = JiRackConfig           # drop-in name compat (optional)
+JiRackTransformer = JiRackTransformer
+JiRackConfig = JiRackConfig          # lets ds7b-style imports work
 JiRackTransformer = JiRackTransformer
 # =============================================================================
+# Smoke test:  python JiRackTernaryPyTorch_ds1p5b.py   (tiny config, CPU, seconds)
 # =============================================================================
 if __name__ == "__main__":
     class TinyConfig(JiRackConfig):
     # 4) lambda survives a state_dict round-trip.
     sd = model.state_dict()
+    model2 = JiRackTransformerDS1p5B(TinyConfig())
     model2.load_state_dict(sd)
     assert abs(model2.get_lambda() - 1.0) < 1e-9, "lambda_ not serialized!"
     print("lambda serialization: OK")
     print(f"export: {sum('codes' in k for k in exported)} ternary tensors, "
           f"biases preserved, OK")
+    print("\nAll smoke tests passed.")

JiRackUltra_1b.Q6_K.gguf ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:8a7a9a4052ded2800d95b37a0bfc8b00630dc417b0e3410785203bab58238463
+size 1464178592