Issues

• GSPO sequence-parallel (SDP/SP) paths are untested (loss + metrics)

  ServiceNow/Fast-LLM#556 · Filter by author jlamypoirier opened on Jul 8, 2026

• Validation loss via LossEvaluator logs 0 (head skips loss computation in eval mode)

  ServiceNow/Fast-LLM#538 · Filter by author jlamypoirier opened on Jun 11, 2026

• Engine support for concatenated weights with per-sub-parameter configuration

  enhancement
  ServiceNow/Fast-LLM#533 · Filter by author jlamypoirier opened on Jun 4, 2026

• Lazy micro-batch input preparation in the schedule runner

  ServiceNow/Fast-LLM#510 · Filter by author jlamypoirier opened on May 6, 2026
  1

• Monolithic fused head-loss kernel for combinable losses + metrics

  ServiceNow/Fast-LLM#507 · Filter by author jlamypoirier opened on May 5, 2026

• feat: add tensor-parallel entropy loss benchmark

  ServiceNow/Fast-LLM#500 · Filter by author jlamypoirier opened on Apr 29, 2026

• perf: Triton normalization backward underperforms apex at wide hidden dimensions

  ServiceNow/Fast-LLM#499 · Filter by author jlamypoirier opened on Apr 29, 2026

• [bug] Can't compile varlen mamba with base image 25.11

  bug
  ServiceNow/Fast-LLM#416 · Filter by author jlamypoirier opened on Dec 11, 2025
  2

• Add torch fallbacks for KDA and GDN layers.

  Feature

  ServiceNow/Fast-LLM#406 · Filter by author oleksost opened on Dec 8, 2025

• [bug] Unable to freeze specific layers of a pretrained model

  bug
  ServiceNow/Fast-LLM#394 · Filter by author RaymondLi0 opened on Nov 25, 2025
  2

  

• [bug] Possible Conversion issue with Hybrid Mamba 2

  bugCritical
  ServiceNow/Fast-LLM#380 · Filter by author jlamypoirier opened on Oct 17, 2025

• Reference-model is slow on long sequences, especially with TP>1

  need update
  ServiceNow/Fast-LLM#353 · Filter by author RaymondLi0 opened on Aug 14, 2025
  2

• Add test for MTP-add-heads script

  Task

  ServiceNow/Fast-LLM#300 · Filter by author RaymondLi0 opened on Jun 12, 2025
  

• Reduce Memory Usage During Inference

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#298 · Filter by author bigximik opened on Jun 12, 2025
  2

• Evaluate Command: Config Design and Functionality for Non-Training Model Evaluation

  enhancementneed update
  Task

  ServiceNow/Fast-LLM#285 · Filter by author bigximik opened on Jun 4, 2025

• Support additional evaluation frameworks

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#283 · Filter by author bigximik opened on Jun 2, 2025
  1

• Support generation with caching for SSMs and SSM–Transformer hybrids.

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#280 · Filter by author bigximik opened on Jun 2, 2025
  1

• Speed up inference performance for attention and ssm layers with caching

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#279 · Filter by author bigximik opened on Jun 2, 2025
  

• Support sequence data parallelism during generation

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#278 · Filter by author bigximik opened on Jun 2, 2025
  1

• [bug] Generate test occasionally fails

  bugneed update
  Bug

  ServiceNow/Fast-LLM#274 · Filter by author bigximik opened on May 22, 2025
  

• Support chat template in prepare

  enhancement
  Feature

  ServiceNow/Fast-LLM#262 · Filter by author sohamparikh opened on May 12, 2025
  

• Elastic training

  enhancement
  Feature

  ServiceNow/Fast-LLM#241 · Filter by author tscholak opened on Apr 24, 2025
  8

  

• Simplify config validation

  enhancementhelp wanted
  ServiceNow/Fast-LLM#232 · Filter by author jlamypoirier opened on Apr 15, 2025
  2

• Reset Number of Steps in WandB to the Latest Saved Checkpoint or Implement Distinguishable Experiment Run Logging in WandB

  enhancementneed update
  Feature

  ServiceNow/Fast-LLM#215 · Filter by author bigximik opened on Mar 31, 2025

• Move Config Validations (e.g., Dataset Usage vs. Definitions) to _validate for Dry Run Checks

  enhancementneed update
  Task

  ServiceNow/Fast-LLM#213 · Filter by author bigximik opened on Mar 28, 2025
  1