Add Support for DeepSeek LLM Integration

[danielmerja]

Description:

DeepSeek is an advanced open-source LLM that rivals some of the most sophisticated models globally, such as OpenAI's offerings. Its latest version, DeepSeek-V3, delivers exceptional inference speed and accuracy across multiple benchmarks, including English, Chinese, code, and math reasoning tasks. Integrating DeepSeek into LangChainGo would expand its capabilities and provide users with an additional high-performance, open-source LLM option.

Current Behavior

LangChainGo currently supports various LLMs but lacks integration with DeepSeek, which is rapidly gaining recognition for its performance and open-source nature.

Expected Behavior

Adding DeepSeek support would allow LangChainGo users to:

1. Leverage DeepSeek's superior inference speed and performance.
2. Access its broad capabilities across multiple domains, such as reasoning, math, and code.
3. Use a free, open-source alternative to proprietary models.

Why This Enhancement Is Useful

1. Performance Leader: DeepSeek ranks highly on key benchmarks like MMLU and HumanEval, outperforming other open-source models and competing with closed-source solutions.
2. Open Source: Aligns with the ethos of LangChainGo by providing more open-source alternatives.
3. Multi-Domain Strength: Its versatility in handling diverse tasks makes it a valuable addition to the LangChainGo ecosystem.

References

• DeepSeek Documentation
• DeepSeek Benchmarks

[lifejwang11]

It use with OpenAI starndard. @danielmerja

[danielmerja]

It use with OpenAI starndard. @danielmerja

Is the finish_reason also supported in the llms/openai?

[lifejwang11]

oh,I agree

[danielmerja]

oh,I agree

it would be very cost effective if we add these, and we can use DeepSeek API with their new reasoning model

Reasoning Model (deepseek-reasoner)

[lifejwang11]

Its a good feature.Every model have their own feature.I think it will support in the future.

[vMaroon]

Any progress on this?

[lifejwang11]

@vMaroon no,I don't have any time for this.

[semioz]

@danielmerja turns out finish_reason was already supported in the existing codebase but not the ReasoningContent. I quickly opened a PR let me know if i missed anything: https://github.com/tmc/langchaingo/pull/1121/files

[danielmerja]

@danielmerja turns out finish_reason was already supported in the existing codebase but not the ReasoningContent. I quickly opened a PR let me know if i missed anything: https://github.com/tmc/langchaingo/pull/1121/files

Sweet! Thank you! I have pinged @FluffyKebab to take a look at it.