Search results

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

Status: Open.

#538 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Jun 11, 2026

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

enhancementNew feature or requestNew feature or request

Status: Open.

#533 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Jun 4, 2026

Lazy micro-batch input preparation in the schedule runner

Status: Open.

#510 In ServiceNow/Fast-LLM;

· jlamypoirier opened on May 6, 2026

Monolithic fused head-loss kernel for combinable losses + metrics

Status: Open.

#507 In ServiceNow/Fast-LLM;

· jlamypoirier opened on May 5, 2026

feat: add tensor-parallel entropy loss benchmark

Status: Open.

#500 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Apr 29, 2026

perf: Triton normalization backward underperforms apex at wide hidden dimensions

Status: Open.

#499 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Apr 29, 2026

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

bugSomething isn't workingSomething isn't working

Status: Open.

#416 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Dec 11, 2025

Add torch fallbacks for KDA and GDN layers.

Status: Open.

Feature

#406 In ServiceNow/Fast-LLM;

· oleksost opened on Dec 8, 2025

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

bugSomething isn't workingSomething isn't working

Status: Open.

#394 In ServiceNow/Fast-LLM;

· RaymondLi0 opened on Nov 25, 2025

[bug] Possible Conversion issue with Hybrid Mamba 2

bugSomething isn't workingSomething isn't working

Critical

Status: Open.

#380 In ServiceNow/Fast-LLM;

· jlamypoirier opened on Oct 17, 2025

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

need update

Status: Open.

#353 In ServiceNow/Fast-LLM;

· RaymondLi0 opened on Aug 14, 2025

Add test for MTP-add-heads script

Status: Open.

Task

#300 In ServiceNow/Fast-LLM;

· RaymondLi0 opened on Jun 12, 2025