FLAMIA: FLAsh MIcroscaling Attention for Fast and Causally Safe Training on Blackwell
READ THE ORIGINAL SOURCEOpen original sourcehuggingface.co · Opens in a new tabFLAMIA is a family of attention kernels for NVIDIA Blackwell GPUs with MXFP8 and mixed NVFP4/FP8 training paths. It combines token-block scaling that reuses quantized operands across matrix orientations, safeguards against causal leakage through shared scales, and consistent forward and backward numerics. Fused quantization and document-local scaling support variable-length packed documents without repacking. The authors report 1.06–1.42× faster combined forward/backward attention on causal packed-document workloads and state that training stability was validated on pretraining Llama-style models.
One primary source; other links are context, not independent corroboration.
- Primary sourceHugging Face Daily Papers
- Discovery linkHugging Face
- Linked in sourceView arXiv page
- Linked in sourceView PDF
The paper page is linked. Peer review and findings have not been independently checked by AINewsIQ.
Open full article view →- Original paper+45
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
FLAMIA is a family of attention kernels for NVIDIA Blackwell GPUs with MXFP8 and mixed NVFP4/FP8 training paths. It combines token-block scaling that reuses quantized operands across matrix orientations, safeguards against causal leakage through shared scales, and consistent forward and backward numerics. Fused quantization and document-local scaling support variable-length packed documents without repacking. The authors report 1.06–1.42× faster combined forward/backward attention on causal packed-document workloads and state that training stability was validated on pretraining Llama-style models.
A READER'S EVIDENCE CHECKHow to read this signalSix questions about claims, sources, risks, and open checks
This is a signal-level guide until claim-level evidence is available. “Not recorded” means the evidence has not been assessed here; it does not mean a result is false. How we check AI news →
- 01Who is making each claim?
The claimant and whether each claim was measured or asserted are not recorded in this public brief.
- 02What was actually tested?
Not recorded per claim. Check the source for sample size, tasks, model and benchmark versions, and reproducibility.
- 03Where are the sources?
Original link: huggingface.co. Brief basis: fetched source page. Primary page date and quote locations are not established in this panel.
- 04What could mislead?
Speedups are reported relative to the fastest evaluated baseline for each workload, and source claims not independently checked.
- 05What is at risk if wrong?
Not assessed per claim. Consider safety, privacy, cost, and adoption decisions before relying on the result.
- 06What would change the view?
No claim-specific check is recorded. An inspectable primary method or independent replication could change the assessment.
Key findings from the stored brief
- The MXFP8 path accelerates combined forward and backward attention by 1.10–1.40× on B200 and 1.06–1.42× on B300 versus the fastest evaluated baseline.
- Shared quantization scales can leak future-token information despite causal masking; local scales and per-key scale folding are used to mitigate this.
- All-FP4 training adds quantization overhead and slightly worsens loss, so the design uses selective low precision instead.
- High single-step gradient similarity alone does not guarantee stable training, according to the paper.
Speedups are reported relative to the fastest evaluated baseline for each workload, and source claims not independently checked.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
The paper page is linked. Peer review and findings have not been independently checked by AINewsIQ.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
Attention is a major cost in long-context Transformer training. The paper shows how quantization design, causal safety, and document scheduling can jointly turn low-bit hardware throughput into practical speedups on Blackwell GPUs.
Limits and open questions
Speedups are reported relative to the fastest evaluated baseline for each workload, and source claims not independently checked.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.