Search results

Release speculators used for ablations

Status: Open.

#57 In deepseek-ai/DeepSpec;

· sheelfshah opened on Jul 7, 2026

[RFC] White-Box Architectural Reference for DSpark: Spherical Normalization, 3-Tier Semantic Cache & Unified Multimodal Token

Status: Open.

#52 In deepseek-ai/DeepSpec;

· Xuan-yi-yan opened on Jul 6, 2026

reading my chats not funny

Status: Open.

#51 In deepseek-ai/DeepSpec;

· ezydubs opened on Jul 4, 2026

一个关于生成范式的远期思考（非Bug报告）

Status: Open.

#49 In deepseek-ai/DeepSpec;

· madheu opened on Jul 3, 2026

RFC: Support Domino speculative decoding in DeepSpec?

Status: Open.

#46 In deepseek-ai/DeepSpec;

· jianuo-huang opened on Jul 2, 2026

launch_sglang_server.sh ignores CUDA_VISIBLE_DEVICES and always launches GPUs 0-7

Status: Open.

#36 In deepseek-ai/DeepSpec;

· Oxygen56 opened on Jul 1, 2026

Concurrent training data generation can corrupt resume order and buffer unbounded results

Status: Open.

#35 In deepseek-ai/DeepSpec;

· morluto opened on Jun 30, 2026

DSpark train/loss metric is rank-weighted instead of globally weighted

Status: Open.

#34 In deepseek-ai/DeepSpec;

· morluto opened on Jun 30, 2026

Eagle3 fallback attention mask crashes with KV cache

Status: Open.

#33 In deepseek-ai/DeepSpec;

· morluto opened on Jun 30, 2026

Warmup to cosine scheduler handoff can overshoot peak LR

Status: Open.

#32 In deepseek-ai/DeepSpec;

· morluto opened on Jun 30, 2026

Draft sampling uses a different distribution than speculative rejection

Status: Open.

#31 In deepseek-ai/DeepSpec;

· morluto opened on Jun 30, 2026

feat: add support for FP8 target cache quantization

Status: Open.

#26 In deepseek-ai/DeepSpec;

· ApurveKaranwal opened on Jun 30, 2026