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 #4717] phi3:medium-128k doesn't use the full context window by default #2974

New Issue
Closed
opened 2026-04-12 13:21:19 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 13:21:19 -05:00
Owner
Copy Link

Originally created by @derluke on GitHub (May 30, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4717

I was playing with the new phi3:medium-128k model and was surprised to see it struggled to keep track of my earlier questions, or handle long documents. But on the bright side it was surprisingly fast.

After a little digging I found out how to specify the context size using a new model file. I decided to give it a go and used this

FROM phi3:medium-128k
TEMPLATE "{{ if .System }}<|system|>
{{ .System }}<|end|>
{{ end }}{{ if .Prompt }}<|user|>
{{ .Prompt }}<|end|>
{{ end }}<|assistant|>
{{ .Response }}<|end|>"
PARAMETER stop <|end|>
PARAMETER stop <|user|>
PARAMETER stop <|assistant|>
PARAMETER num_ctx 65536

(I wasn't sure exactly how much 128k is supposed to be (assume 2**17) so decided to be on the safe side and take the power of two below)
And it worked, now the model can read long documents and behaves as expected (and is rather slow, but that is due to my limited hardware)

It would be amazing if these models came pre-configured such that they use the full context window by default.

Originally created by @derluke on GitHub (May 30, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4717 I was playing with the new phi3:medium-128k model and was surprised to see it struggled to keep track of my earlier questions, or handle long documents. But on the bright side it was surprisingly fast. After a little digging I found out how to specify the context size using a new model file. I decided to give it a go and used this ``` FROM phi3:medium-128k TEMPLATE "{{ if .System }}<|system|> {{ .System }}<|end|> {{ end }}{{ if .Prompt }}<|user|> {{ .Prompt }}<|end|> {{ end }}<|assistant|> {{ .Response }}<|end|>" PARAMETER stop <|end|> PARAMETER stop <|user|> PARAMETER stop <|assistant|> PARAMETER num_ctx 65536 ``` (I wasn't sure exactly how much 128k is supposed to be (assume 2**17) so decided to be on the safe side and take the power of two below) And it worked, now the model can read long documents and behaves as expected (and is rather slow, but that is due to my limited hardware) It would be amazing if these models came pre-configured such that they use the full context window by default.
GiteaMirror added the feature request label 2026-04-12 13:21:19 -05:00
GiteaMirror commented 2026-04-12 13:21:22 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (May 30, 2024):

Thanks for the issue! Will close this for https://github.com/ollama/ollama/issues/1005

<!-- gh-comment-id:2140142972 --> @jmorganca commented on GitHub (May 30, 2024): Thanks for the issue! Will close this for https://github.com/ollama/ollama/issues/1005
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 feature request
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#2974
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?