Add MiniMax as a first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a first-class LLM provider in Langfun, following the existing patterns used by DeepSeek and Groq providers.

Changes

• langfun/core/llms/minimax.py — New provider module:

  • MiniMaxModelInfo extending lf.ModelInfo with provider metadata and links
  • MiniMax base class extending OpenAIChatCompletionAPI with MiniMax-specific configuration
  • MiniMaxM27 and MiniMaxM27Highspeed convenience classes for MiniMax-M2.7 and MiniMax-M2.7-highspeed models
  • Temperature clamping (MiniMax API requires temperature ∈ (0.0, 1.0], so 0.0 is clamped to 0.01)
  • API key management via api_key parameter or MINIMAX_API_KEY environment variable
  • Model registration with lf.LanguageModel.register() for LanguageModel.get() support

• langfun/core/llms/minimax_test.py — 10 unit tests covering:

  • Model directory listing
  • API key validation and environment variable fallback
  • Model ID and resource ID verification
  • Model info metadata
  • Request formatting
  • Temperature clamping behavior
  • LanguageModel.get() registration

• langfun/core/llms/__init__.py — Added MiniMax exports

• README.md — Added MiniMax to the list of supported LLMs

Supported Models

Model	Context Window	Description
MiniMax-M2.7	1M tokens	Latest flagship model
MiniMax-M2.7-highspeed	1M tokens	Optimized for speed

Usage

import langfun as lf

# Using convenience class
lm = lf.llms.MiniMaxM27(api_key='YOUR_KEY')
# Or via environment variable MINIMAX_API_KEY
lm = lf.llms.MiniMaxM27()

r = lm('Who are you?')
print(r)

# Using LanguageModel.get()
lm = lf.LanguageModel.get('MiniMax-M2.7')

Test Results

All 10 unit tests pass. Existing provider tests (DeepSeek, Groq) continue to pass without regressions.

---

4 files changed, 288 additions, 1 deletion

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[JiwaniZakir]

In _register_minimax_models(), both model IDs are registered against the base MiniMax class rather than their respective typed subclasses (MiniMaxM27, MiniMaxM27Highspeed). This means lf.LanguageModel.make('MiniMax-M2.7') would return a base MiniMax instance that still requires the model parameter to be set, rather than a pre-configured subclass instance — inconsistent with how other providers like DeepSeek wire up their registrations.

Additionally, the _request_args override in MiniMax clamps temperature == 0.0 to 0.01 since the API requires (0.0, 1.0], but there's no guard on the upper bound — if a caller passes temperature > 1.0, it will be forwarded as-is and likely cause a 4xx from the MiniMax API. The other boundary should either be clamped or raise a clear ValueError at validation time.

Minor: the blank line between _register_minimax_models() definition and its module-level call is missing (line 179), which is inconsistent with PEP 8 style used elsewhere in the repo.