Unable to use ollama create to load custom gguf model

[Ng Warren]

Issue: Error When Using ollama create to Load a Model

Description

When attempting to use the ollama create API to load a model, the following error is returned:

*** Error:
{"error":"path or Modelfile are required"}
***

Environment

• Ollama Version: 0.5.4
• Docker Compose Service: ollama running on localhost:8015

---

Steps to Reproduce

1. Generate SHA-256 Checksum:

   sha256sum {model path}

   Example:

   sha256sum Llama-3.2-3B-Instruct-Q4_K_M.gguf

   Output:

   6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff  Llama-3.2-3B-Instruct-Q4_K_M.gguf
   

2. Push Blob to Ollama:

   curl -T Llama-3.2-3B-Instruct-Q4_K_M.gguf -X POST http://localhost:8015/api/blobs/sha256:6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff

3. Verify Blob Exists:

   curl -I http://localhost:8015/api/blobs/sha256:6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff

   Output:

   HTTP/1.1 200 OK
   Date: Thu, 03 Apr 2025 06:43:44 GMT
   

4. Use ollama create to Load the Model:

   curl http://localhost:8015/api/create -d '{
     "model": "onepiece",
     "files": {
       "Llama-3.2-3B-Instruct-Q4_K_M.gguf": "sha256:6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff"
     },
     "template": "{{- if .System }}\n<|system|>\n{{ .System }}\n</s>\n{{- end }}\n<|user|>\n{{ .Prompt }}\n</s>\n<|assistant|>",
     "parameters": {
         "temperature": 0.2,
         "num_ctx": 8192,
         "stop": ["<|system|>", "<|user|>", "<|assistant|>", "</s>"]
     },
     "system": "You are Luffy from One Piece, acting as an assistant."
   }'

5. Error Output:

   {"error":"path or Modelfile are required"}
   

---

Expected Behavior

The model should be successfully created and loaded into the Ollama server.

---

Actual Behavior

The server returns the following error:

{"error":"path or Modelfile are required"}

---

Additional Information

• Blob Verification Screenshot:
  

• Command Used:

  curl http://localhost:8015/api/create -d '{
    "model": "onepiece",
    "files": {
      "Llama-3.2-3B-Instruct-Q4_K_M.gguf": "sha256:6c1a2b41161032677be168d354123594c0e6e67d2b9227c84f296ad037c728ff"
    },
    "template": "{{- if .System }}\n<|system|>\n{{ .System }}\n</s>\n{{- end }}\n<|user|>\n{{ .Prompt }}\n</s>\n<|assistant|>",
    "parameters": {
        "temperature": 0.2,
        "num_ctx": 8192,
        "stop": ["<|system|>", "<|user|>", "<|assistant|>", "</s>"]
    },
    "system": "You are Luffy from One Piece, acting as an assistant."
  }'

• Model Reference:
  The model used for this process is available at Llama-3.2-3B-Instruct-Q4_K_M.gguf.

---

Questions

1. Is the path or modelfile field required in the /api/create payload? If so, how should it be structured?
2. Are there additional steps or configurations needed to successfully create a model from a GGUF file?

---

Request for Assistance

Any guidance on how to resolve this issue and successfully create a model using the /api/create endpoint is greatly appreciated!!!

OS

Docker

GPU

Intel

CPU

Intel

Ollama version

0.5.4

[Kai Huang]

Hi, the request format you are using is for recent versions of ollama, but for ollama 0.5.4 it is not supported: https://github.com/ollama/ollama/blob/v0.5.4/api/types.go#L297
Seems for 0.5.4 you can only create using a Modelfile following the sample command here: https://github.com/ollama/ollama/blob/v0.5.4/docs/api.md#request-20

We are actively upgrading ollama to v0.6.x and you can use your command after we complete the upgrade. Will keep you updated about our upgrade progress, thanks!

[Ng Warren]

Alright thanks for the information. Cheers!