github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 13:19:40 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #3110] LLaMA2/Mistral generation extremely slow and errorenous #1912

New Issue
Closed
opened 2026-04-12 12:02:00 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 12:02:00 -05:00
Owner
Copy Link

Originally created by @RoManInv on GitHub (Mar 13, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/3110

Originally assigned to: @jmorganca on GitHub.

Hi all,
I just experienced a situation where LLaMA2 and Mistral generate a completion very slowly (sometimes even no response after around 1h) and seem not to be correct. I would like to ask if there is anything wrong when I use the kwarg "format": "json" in the request body, or if it is my prompt that caused that issue.

  • Server spec
    • CPU: enterprise-level 256 core
    • GPU: multiple Nvidia A-100 GPUs
    • System: Debian Bullseye
    • Access: admin, full access to all the computational resources
    • Others: ollama was the only process running on that server. Ollama is installed on the host machine, not VM or Docker. There is no VM/Docker or anything that restricts the computational resource for any user.

Here is the test case I'm using:

  • Preparation:

    • sudo service ollama start --> success
    • ollama pull llama2 --> success
    • ollama pull mistral --> success

  • LLaMA2 test call:

curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
  "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:",
  "stream": false,
  "format": "json",
  "options": {
    "seed": 101,
    "temperature": 0.2
  }
}'
  • Response:
{"model":"llama2","created_at":"2024-03-13T14:55:18.825061607Z","response":"{ }\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n","done":true,"context":[518,25580,29962,3532,14816,29903,29958,5299,829,14816,29903,6778,13,13,17813,278,1494,4766,29889,660,29901,7551,319,29901,7551,29892,1522,823,292,660,29901,5115,319,29901,9556,29892,5115,660,29901,12730,319,29901,518,29914,25580,29962,13,29912,500,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13],"total_duration":16179381781,"load_duration":610419,"prompt_eval_duration":20061000,"eval_count":1001,"eval_duration":16157638000}%

However, if I remove the format keyword, it is a little bit faster with some response seems to be reasonable.

  • LLaMA test call removing "format": "json" in the request body:
curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
  "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:",
  "stream": false,

  "options": {
    "seed": 101,
    "temperature": 0.2
  }
}'
  • Response of the above request:
{"model":"llama2","created_at":"2024-03-13T15:01:10.874304486Z","response":"\nSure! Here's the completed pattern:\n\nQ: China A: China, Beijing\nQ: Berlin A: Germany, Berlin\nQ: Italy A: Rome","done":true,"context":[518,25580,29962,3532,14816,29903,29958,5299,829,14816,29903,6778,13,13,17813,278,1494,4766,29889,660,29901,7551,319,29901,7551,29892,1522,823,292,660,29901,5115,319,29901,9556,29892,5115,660,29901,12730,319,29901,518,29914,25580,29962,13,13,29903,545,29991,2266,29915,29879,278,8676,4766,29901,13,13,29984,29901,7551,319,29901,7551,29892,1522,823,292,13,29984,29901,5115,319,29901,9556,29892,5115,13,29984,29901,12730,319,29901,9184],"total_duration":1596398280,"load_duration":1102959851,"prompt_eval_count":48,"prompt_eval_duration":122206000,"eval_count":40,"eval_duration":370584000}%

For Mistral, the situation is worse. It won't give response in around 1 hour.

  • Mistral test call:
curl http://localhost:11434/api/generate -d '{
  "model": "mistral",
  "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:",
  "stream": false,
  "format": "json",
  "options": {
    "seed": 101,
    "temperature": 0.2
  }
}'

  • Response: No response after around an hour, call aborted by sending Ctrl-C command. During waiting, htop says ollama was running and actually taking computational resources.

  • Mistral test call removing "format": "json" in the request body:

curl http://localhost:11434/api/generate -d '{
  "model": "mistral",
  "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:",
  "stream": false,

  "options": {
    "seed": 101,
    "temperature": 0.2
  }
}'
  • Response (seems reasonable):
{"model":"mistral","created_at":"2024-03-13T15:12:29.747354811Z","response":" Italy, Rome\n\nThe pattern is to list the country name first, followed by a specific city that is well-known in that country. In this case, for Italy, the city \"Rome\" was given as an example.","done":true,"context":[733,16289,28793,28705,21929,272,2296,5340,28723,1186,28747,5077,330,28747,5077,28725,27112,1186,28747,8430,330,28747,7293,28725,8430,1186,28747,10828,330,28747,733,28748,16289,28793,10828,28725,10827,13,13,1014,5340,349,298,1274,272,2939,1141,907,28725,4961,486,264,2948,2990,369,349,1162,28733,4717,297,369,2939,28723,560,456,1222,28725,354,10828,28725,272,2990,345,28754,525,28739,403,2078,390,396,2757,28723],"total_duration":1943432265,"load_duration":1335220122,"prompt_eval_count":35,"prompt_eval_duration":134002000,"eval_count":49,"eval_duration":473637000}%

Additionally, the examples provided in the API documentation worked fine with and without "format": "json". A response is given in a very short time.

Originally created by @RoManInv on GitHub (Mar 13, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/3110 Originally assigned to: @jmorganca on GitHub. Hi all, I just experienced a situation where LLaMA2 and Mistral generate a completion very slowly (sometimes even no response after around 1h) and seem not to be correct. I would like to ask if there is anything wrong when I use the kwarg "format": "json" in the request body, or if it is my prompt that caused that issue. * Server spec - CPU: enterprise-level 256 core - GPU: multiple Nvidia A-100 GPUs - System: Debian Bullseye - Access: admin, full access to all the computational resources - Others: ollama was the only process running on that server. Ollama is installed on the host machine, not VM or Docker. There is no VM/Docker or anything that restricts the computational resource for any user. Here is the test case I'm using: * Preparation: - sudo service ollama start --> success - ollama pull llama2 --> success - ollama pull mistral --> success * LLaMA2 test call: ``` curl http://localhost:11434/api/generate -d '{ "model": "llama2", "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:", "stream": false, "format": "json", "options": { "seed": 101, "temperature": 0.2 } }' ``` * Response: ``` {"model":"llama2","created_at":"2024-03-13T14:55:18.825061607Z","response":"{ }\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n","done":true,"context":[518,25580,29962,3532,14816,29903,29958,5299,829,14816,29903,6778,13,13,17813,278,1494,4766,29889,660,29901,7551,319,29901,7551,29892,1522,823,292,660,29901,5115,319,29901,9556,29892,5115,660,29901,12730,319,29901,518,29914,25580,29962,13,29912,500,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13],"total_duration":16179381781,"load_duration":610419,"prompt_eval_duration":20061000,"eval_count":1001,"eval_duration":16157638000}% ``` However, if I remove the format keyword, it is a little bit faster with some response seems to be reasonable. * LLaMA test call removing "format": "json" in the request body: ``` curl http://localhost:11434/api/generate -d '{ "model": "llama2", "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:", "stream": false, "options": { "seed": 101, "temperature": 0.2 } }' ``` * Response of the above request: ``` {"model":"llama2","created_at":"2024-03-13T15:01:10.874304486Z","response":"\nSure! Here's the completed pattern:\n\nQ: China A: China, Beijing\nQ: Berlin A: Germany, Berlin\nQ: Italy A: Rome","done":true,"context":[518,25580,29962,3532,14816,29903,29958,5299,829,14816,29903,6778,13,13,17813,278,1494,4766,29889,660,29901,7551,319,29901,7551,29892,1522,823,292,660,29901,5115,319,29901,9556,29892,5115,660,29901,12730,319,29901,518,29914,25580,29962,13,13,29903,545,29991,2266,29915,29879,278,8676,4766,29901,13,13,29984,29901,7551,319,29901,7551,29892,1522,823,292,13,29984,29901,5115,319,29901,9556,29892,5115,13,29984,29901,12730,319,29901,9184],"total_duration":1596398280,"load_duration":1102959851,"prompt_eval_count":48,"prompt_eval_duration":122206000,"eval_count":40,"eval_duration":370584000}% ``` For Mistral, the situation is worse. It won't give response in around 1 hour. * Mistral test call: ``` curl http://localhost:11434/api/generate -d '{ "model": "mistral", "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:", "stream": false, "format": "json", "options": { "seed": 101, "temperature": 0.2 } }' ``` * Response: No response after around an hour, call aborted by sending Ctrl-C command. During waiting, htop says ollama was running and actually taking computational resources. * Mistral test call removing "format": "json" in the request body: ``` curl http://localhost:11434/api/generate -d '{ "model": "mistral", "prompt":"Complete the following pattern. Q: China A: China, Beijing Q: Berlin A: Germany, Berlin Q: Italy A:", "stream": false, "options": { "seed": 101, "temperature": 0.2 } }' ``` * Response (seems reasonable): ``` {"model":"mistral","created_at":"2024-03-13T15:12:29.747354811Z","response":" Italy, Rome\n\nThe pattern is to list the country name first, followed by a specific city that is well-known in that country. In this case, for Italy, the city \"Rome\" was given as an example.","done":true,"context":[733,16289,28793,28705,21929,272,2296,5340,28723,1186,28747,5077,330,28747,5077,28725,27112,1186,28747,8430,330,28747,7293,28725,8430,1186,28747,10828,330,28747,733,28748,16289,28793,10828,28725,10827,13,13,1014,5340,349,298,1274,272,2939,1141,907,28725,4961,486,264,2948,2990,369,349,1162,28733,4717,297,369,2939,28723,560,456,1222,28725,354,10828,28725,272,2990,345,28754,525,28739,403,2078,390,396,2757,28723],"total_duration":1943432265,"load_duration":1335220122,"prompt_eval_count":35,"prompt_eval_duration":134002000,"eval_count":49,"eval_duration":473637000}% ``` Additionally, the examples provided in the API documentation worked fine with and without "format": "json". A response is given in a very short time.
GiteaMirror added the bugperformance labels 2026-04-12 12:02:00 -05:00
GiteaMirror commented 2026-04-12 12:02:03 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Nov 15, 2024):

This should perform much better with newer models (i.e. llama3.1 or llama3.2). Keep in mind when using JSON mode to specify in the prompt that the reply should be JSON

<!-- gh-comment-id:2479475135 --> @jmorganca commented on GitHub (Nov 15, 2024): This should perform much better with newer models (i.e. `llama3.1` or `llama3.2`). Keep in mind when using JSON mode to specify in the prompt that the reply should be JSON
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug performance
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#1912
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?