brb-nv

feat: Support prequantized fp8 ckpt for nemotron-mini-4b-instruct

12

This MR adds changes to support consumption of pre-quantized FP8 ckpt from ModelOpt for nemotron-mini-4b-instruct. ``` $ rm -rf nemotron_mini_4b_tllm_fp8_ckpt/ && python examples/gpt/convert_checkpoint.py --model_dir nemotron-mini-4b-instruct_vfp8-fp8-bf16-export/ --output_dir nemotron_mini_4b_tllm_fp8_ckpt/ --dtype bfloat16 $...

feat:Add support for Phi-4-mini and Phi-4-MM

4

This MR adds support for Phi-4-mini and Phi-4-multimodal models.

test:Add Eagle tests with untrained heads

10

This MR adds unit tests to validate Eagle support for models with untrained Eagle heads. These are meant to be sanity tests which will find blatant issues such as missing...

test: Add tests for Ministral, Codestral, Mistral Small

2

This MR adds tests validating FP8, LoRA, Medusa support for following models: 1) Codestral-22B-v0.1 2) Ministral-8B-Instruct-2410 3) Mistral-Small-24B-Base-2501

[None][feat] Integrate helix parallelism

1

## Description This MR integrates helix parallelism, an experimental feature, in TRTLLM. **_Background:_** - Helix parallelism is a decode-only context parallelism method. Hence, it's used in disaggregated setting where only...