feat: add Fireworks AI component

[namwoam]

Because

• We want to integrate Fireworks AI Client into our VDP pipeline platform.

This commit

• Added the Fireworks AI Component, which supports the following tasks:
  • (a) TASK_TEXT_GENERATION_CHAT
    Models: llama3.1 (405b, 70b, 9b), llama3 (70b, 8b), gemma2-9b, phi-3-vision-128K, deepseek-coder, qwen2 (72b) and more
  • (b) TASK_TEXT_EMBEDDINGS
    Models: nomic-ai/nomic-embed, thenlper/gte, WhereIsAI/UAE

API Documentation: https://docs.fireworks.ai/api-reference/introduction
Pricing Documentation: https://fireworks.ai/pricing

[namwoam]

Above is the performance of llama3-70b-it across different providers. This sector is extremely competitive and I am not sure if we will support all the providers?

[codecov]

Codecov Report

Attention: Patch coverage is 55.80110% with 80 lines in your changes missing coverage. Please review.

Project coverage is 37.15%. Comparing base (4aeae69) to head (21e4fb7).
Report is 20 commits behind head on main.

Files	Patch %	Lines
ai/fireworksai/v0/tasks.go	55.67%	38 Missing and 5 partials ⚠️
ai/fireworksai/v0/client.go	36.00%	16 Missing ⚠️
ai/fireworksai/v0/main.go	71.69%	13 Missing and 2 partials ⚠️
store/store.go	0.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #237      +/-   ##
==========================================
- Coverage   40.70%   37.15%   -3.55%     
==========================================
  Files         121      153      +32     
  Lines       11842    19404    +7562     
==========================================
+ Hits         4820     7210    +2390     
- Misses       6402    11138    +4736     
- Partials      620     1056     +436     

Flag	Coverage Δ
unittests	37.15% <55.80%> (-3.55%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[chuang8511]

lgtm.

[chuang8511]

@namwoam
Btw, does Fireworks have their official tokenizer that we can use?
It will be used in Text Component ChunkTask

[namwoam]

All their model are open sourced, so I guess it could be imported from huggingface?

[chuang8511]

@namwoam
Could you help me check the pricing?
I am not sure these models' params count.

remaining models

deepseek-coder-v2-instruct
deepseek-coder-v2-lite-instruct
phi-3-vision-128k-instruct
nomic-ai/nomic-embed-text-v1.5
nomic-ai/nomic-embed-text-v1
WhereIsAI/UAE-Large-V1
thenlper/gte-large
thenlper/gte-base

[namwoam]

Hello, the parameter count are as follows:

Text models:

1. deepseek-coder-v2-instruct: 236b MoE
2. deepseek-coder-v2-lite-instruct: 16b MoE source: huggingface
3. phi-3-vision-128k-instruct: 4.2b source: phi3 technical report under section 6.1

Embedding models:

1. nomic-embed-text-v1.5: 137m source: huggingface
2. nomic-embed-text-v1: 137m source: huggingface
3. UAE-Large-V1: 335m source: huggingface
4. gte-large: 335m source: huggingface
5. gte-base: 109m source: huggingface

The pricing for embedding models:

[chuang8511]

@namwoam
I still cannot match which price is correct. I will take out from Instill Credit first.
Please help us check it again. Thank you.

• deepseek-coder-v2-instruct: 236b MoE
  

✅ deepseek-coder-v2-lite-instruct: 16b MoE source: huggingface
-> 0.5
✅ phi-3-vision-128k-instruct: 4.2b source: phi3 technical report under section 6.1

-> 0.2

[namwoam]

Ok, no problem.