github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #2046] What quantization is used to quantize Phi-2? #26942

New Issue
Closed
opened 2026-04-22 03:44:11 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-22 03:44:11 -05:00
Owner
Copy Link

Originally created by @bm777 on GitHub (Jan 18, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2046

Running Phi-2 with Ollama is faster than running Phi-2 in Rust with Candle. rust is taking 1.7 GB of my memory while Ollama only 788MB of memory. I guess it is using the same GGUF quantized 1.6 Gb

Ollama is

  • quantizing it at run time or
  • it does it before hand
  • or using lama.cpp under the hood
  • no quantization at all.
Originally created by @bm777 on GitHub (Jan 18, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2046 Running Phi-2 with Ollama is faster than running Phi-2 in Rust with Candle. rust is taking 1.7 GB of my memory while Ollama only 788MB of memory. I guess it is using the same GGUF quantized 1.6 Gb Ollama is - quantizing it at run time or - it does it before hand - or using lama.cpp under the hood - no quantization at all. ----
GiteaMirror commented 2026-04-22 03:44:13 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (Jan 18, 2024):

Look at https://ollama.ai/library/phi/tags. You can check the fingerprint to figure out which quantization is used for phi:latest. Or not, because at this point you can pretty much count on it being q4_0 for any model in the ollama.ai/library.

That said, your memory utilization figure probably off for Ollama. It uses llama.cpp under the hood and mmaps the model weights. This doesn't show up as part of the processes memory. Instead it's accounted for under the file cache on linux and either wired memory (when an inference is in progress), or file cache (when idle) on MacOS.

<!-- gh-comment-id:1898871984 --> @easp commented on GitHub (Jan 18, 2024): Look at https://ollama.ai/library/phi/tags. You can check the fingerprint to figure out which quantization is used for phi:latest. Or not, because at this point you can pretty much count on it being q4_0 for any model in the ollama.ai/library. That said, your memory utilization figure probably off for Ollama. It uses llama.cpp under the hood and mmaps the model weights. This doesn't show up as part of the processes memory. Instead it's accounted for under the file cache on linux and either wired memory (when an inference is in progress), or file cache (when idle) on MacOS.
GiteaMirror commented 2026-04-22 03:44:14 -05:00
Author
Owner
Copy Link

@bm777 commented on GitHub (Jan 19, 2024):

@easp thanks :)

<!-- gh-comment-id:1899680994 --> @bm777 commented on GitHub (Jan 19, 2024): @easp thanks :)
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
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#26942
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?