Do not set the ctx-size by default #1915
Conversation
rhatdan
requested review from
bmahabirbu,
maxamillion,
swarajpande5,
jhjaggars,
cgruver and
engelmi
as code owners
Reviewer's GuideThis PR replaces the hard-coded context size default with a zero placeholder to allow llama-server’s built-in default and introduces a new cache_reuse configuration and CLI argument, updating runtime invocation, configuration schema, documentation, and tests accordingly. Sequence diagram for llama_serve command argument constructionsequenceDiagram
participant "User/CLI"
participant "ramalama.model.llama_serve()"
participant "llama-server"
"User/CLI"->>"ramalama.model.llama_serve()": Provide --ctx-size and --cache-reuse args
"ramalama.model.llama_serve()"->>"llama-server": Build exec_args
alt context > 0
"ramalama.model.llama_serve()"->>"llama-server": Pass --ctx-size <value>
else context == 0
"ramalama.model.llama_serve()"->>"llama-server": Omit --ctx-size (use llama-server default)
end
"ramalama.model.llama_serve()"->>"llama-server": Pass --cache-reuse <value>
"llama-server"-->>"User/CLI": Use provided or default values
Entity relationship diagram for updated ramalama.conf configuration optionserDiagram
CONFIG {
int ctx_size
int cache_reuse
}
CONFIG ||--|| "ramalama.conf" : defines
Class diagram for updated configuration options in BaseConfigclassDiagram
class BaseConfig {
api: str = "none"
carimage: str = "registry.access.redhat.com/ubi10-micro:latest"
container: bool
ctx_size: int = 0
cache_reuse: int = 256
default_image: str
dryrun: bool
engine: SUPPORTED_ENGINES | None
...
}
Class diagram for updated CLI argument parsing in runtime_optionsclassDiagram
class runtime_options {
+parser.add_argument("--cache-reuse", dest="cache_reuse", type=int, default=CONFIG.cache_reuse)
+parser.add_argument("--ctx-size", "-c", dest="context", type=int, default=CONFIG.ctx_size)
+parser.add_argument("--max-model-len", dest="context", type=int, default=CONFIG.ctx_size)
}
File-Level Changes
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @rhatdan, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request improves the flexibility and default behavior of ramalama when interacting with llama.cpp. It adjusts the handling of the ctx-size parameter to align with llama.cpp's default behavior and introduces user-configurable options for the cache_reuse parameter, enhancing control over model serving.
Highlights
- Context Size Defaulting: The ctx-size parameter for llama.cpp is no longer hardcoded to 2048 but defaults to 0 in ramalama, allowing llama.cpp to use its internal default (4098). The --ctx-size argument is now only passed to llama-server if its value is explicitly set to greater than 0.
- Cache Reuse Configurability: The cache_reuse parameter, previously hardcoded to 256, can now be configured by users via ramalama.conf or command-line arguments, providing more flexibility.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Hey there - I've reviewed your changes and they look great!
Prompt for AI Agents
Please address the comments from this code review:
## Individual Comments
### Comment 1
<location> `ramalama/cli.py:779` </location>
<code_context>
)
parser.add_argument("--authfile", help="path of the authentication file")
if command in ["run", "perplexity", "serve"]:
+ parser.add_argument(
+ "--cache-reuse",
+ dest="cache_reuse",
+ type=int,
+ default=CONFIG.cache_reuse,
+ help="min chunk size to attempt reusing from the cache via KV shifting",
+ completer=suppressCompleter,
+ )
parser.add_argument(
</code_context>
<issue_to_address>
Consider clarifying the unit for --cache-reuse in the help text.
Specifying the unit in the help text will make it clearer for users and prevent misunderstandings.
</issue_to_address>
<suggested_fix>
<<<<<<< SEARCH
help="min chunk size to attempt reusing from the cache via KV shifting",
=======
help="min chunk size (in bytes) to attempt reusing from the cache via KV shifting",
>>>>>>> REPLACE
</suggested_fix>
### Comment 2
<location> `ramalama/model.py:662` </location>
<code_context>
parser.add_argument("--authfile", help="path of the authentication file")
if command in ["run", "perplexity", "serve"]:
+ parser.add_argument(
+ "--cache-reuse",
+ dest="cache_reuse",
+ type=int,
</code_context>
<issue_to_address>
Switching --cache-reuse from a hardcoded value to a parameter increases flexibility but may require validation.
Please add validation for the --cache-reuse parameter to prevent invalid values and potential performance issues.
</issue_to_address>Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.
| dest="cache_reuse", | ||
| type=int, | ||
| default=CONFIG.cache_reuse, | ||
| help="min chunk size to attempt reusing from the cache via KV shifting", |
There was a problem hiding this comment.
suggestion: Consider clarifying the unit for --cache-reuse in the help text.
Specifying the unit in the help text will make it clearer for users and prevent misunderstandings.
| help="min chunk size to attempt reusing from the cache via KV shifting", | |
| help="min chunk size (in bytes) to attempt reusing from the cache via KV shifting", |
| "--temp", | ||
| f"{args.temp}", | ||
| "--cache-reuse", | ||
| "256", |
There was a problem hiding this comment.
suggestion: Switching --cache-reuse from a hardcoded value to a parameter increases flexibility but may require validation.
Please add validation for the --cache-reuse parameter to prevent invalid values and potential performance issues.
There was a problem hiding this comment.
Code Review
This pull request correctly changes the default ctx-size to 0 to use the llama.cpp default and introduces configurability for cache_reuse. The implementation looks solid, with corresponding updates to documentation and tests. I've noted a minor potential inconsistency in the documentation regarding the default context size.
|
|
||
| #### **--ctx-size**, **-c** | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 2048, 0 = loaded from model) | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 4096, 0 = loaded from model) |
There was a problem hiding this comment.
The PR description mentions that llama.cpp defaults to a context size of 4098, but the documentation here states the default is 4096. To ensure accuracy, could you please verify the current default ctx-size in llama.cpp and update the documentation accordingly? This will help avoid confusion for users relying on the default behavior.
|
|
||
| #### **--ctx-size**, **-c** | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 2048, 0 = loaded from model) | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 4096, 0 = loaded from model) |
There was a problem hiding this comment.
The PR description mentions that llama.cpp defaults to a context size of 4098, but the documentation here states the default is 4096. To ensure accuracy, could you please verify the current default ctx-size in llama.cpp and update the documentation accordingly? This will help avoid confusion for users relying on the default behavior.
|
|
||
| #### **--ctx-size**, **-c** | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 2048, 0 = loaded from model) | ||
| size of the prompt context. This option is also available as **--max-model-len**. Applies to llama.cpp and vllm regardless of alias (default: 4096, 0 = loaded from model) |
There was a problem hiding this comment.
The PR description mentions that llama.cpp defaults to a context size of 4098, but the documentation here states the default is 4096. To ensure accuracy, could you please verify the current default ctx-size in llama.cpp and update the documentation accordingly? This will help avoid confusion for users relying on the default behavior.
llama.cpp is defaulting to ctx-size of 4096 and we were hard coding 2048, which means we were not using the default setting. This PR changes to use ctx-size=0 which will not be specified in the command unless the value is > 0, so llama-server default will be used. Also we hard coded cache_reuse=256 with no way for user to override, this PR adds support for cache_reuse being set in ramalama.conf and on the command line. Signed-off-by: Daniel J Walsh <[email protected]>
llama.cpp is defaulting to ctx-size of 4098 and we were hard coding 2049, which means we were not using the default setting. This PR changes to use ctx-size=0 which will not be specified in the command unless the value is > 0, so llama-server default will be used.
Also we hard coded cache_reuse=256 with no way for user to override, this PR adds support for cache_reuse being set in ramalama.conf and on the command line.
Summary by Sourcery
Enable user-configurable cache reuse and switch to using llama-server’s default context size by default instead of hard-coding a ctx_size
New Features:
Enhancements:
Documentation:
Tests: