ViperEkura
0 Followers · 0 Following
  • Joined on 2026-04-02
Repositories
10
 Projects Packages Public Activity Starred Repositories
ViperEkura pushed to main at ViperEkura/AstrAI 2026-05-09 16:00:02 +08:00
ca4e6b907c feat: Checkpoint 支持 extra 通用扩展数据，用户通过函数自定义保存/恢复优化器等状态
db99d8b254 fix: 修复文档多处不准确 + inference scheduler 越界 bug + SchedulerCallback 回调阶段修正
b98c9cefdc refactor: 移除 device_ids 参数设计，统一通过 CUDA_VISIBLE_DEVICES 控制 GPU 分配；更新 README 训练示例
283bcaf2ff fix: 修复 CLI 参数缺失/重复、device_ids 越界、generate 参数名不一致、scheduler 时序、非流式截断等 bug
Compare 4 commits »
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-09 14:35:57 +08:00
c9f290c3c8 Merge PROMO_GUIDE into README, add AGENTS.md, parallelize render_all
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-09 13:53:21 +08:00
f5499866ad refactor: replace prefix cache with paged KV cache across promo guide, architecture, and animation
ViperEkura pushed to main at ViperEkura/AstrAI 2026-05-09 12:25:47 +08:00
bc7c82977e feat: GRPO CLI 接入 + on-policy，OpenAI API top_k 参数化，补充训练参数表
34a511e36e feat: 新增 Docker Compose 一键部署，支持 GPU/CPU 双模式
d73f52a2f8 feat: 新增 Anthropic 兼容 /v1/messages API，移除旧版 /generate 端点
Compare 3 commits »
ViperEkura pushed to main at ViperEkura/AstrAI 2026-05-08 23:01:53 +08:00
9d96b0431d docs: 更新文档以匹配分页 KV cache 等代码重构
f81e2b4a73 feat: OpenAI 兼容的 chat completion API（流式+非流式+usage）
4e324d8f26 fix: benchmark 改用 PagedCache 替代已删除的 persistent_key_values
6ed0506491 fix: 减少调度器延迟 — 移除解码路径 5ms 睡眠，修复 refill 任务丢失 bug
30cc2d67a4 refactor: 分页 KV cache 替换固定 slot，删除 PrefixCache 及相关死代码
Compare 14 commits »
ViperEkura pushed to main at ViperEkura/postgraduate-prep 2026-05-08 18:19:30 +08:00
97af47ecc0 fix: README积分导航改为按要点逐条列出
ViperEkura pushed to main at ViperEkura/postgraduate-prep 2026-05-08 18:17:45 +08:00
5640bb167f refactor: 拆分积分章节为4个子章节，更新README导航
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 23:22:04 +08:00
740e766e99 refactor: remove summary text from prefix_cache, leave bottom space for subtitles
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 23:14:33 +08:00
01ccf192ea refactor: extract CTA scene into standalone cta.py
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 23:08:25 +08:00
c2c18f10d6 fix: replace unicode checkmark with ASCII argmax label
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 22:47:28 +08:00
e594d02d31 feat: merge D,E,F into one batch (gated by F at t=8) to show worse static idle
ViperEkura pushed to main at ViperEkura/SKILLS 2026-05-07 22:33:20 +08:00
97355cefad feat: 添加 \\(...\\) 内联数学公式支持
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 22:25:35 +08:00
29007c7b97 chore: remove T0-T3 demo tokens from position-grouped decode section
ViperEkura pushed to main at ViperEkura/postgraduate-prep 2026-05-07 20:59:12 +08:00
ab308aafab refactor: 统一分式/根号积分公式为a²x²+b²形式
ViperEkura pushed to main at ViperEkura/postgraduate-prep 2026-05-07 20:47:00 +08:00
a9f8e6abc5 feat: 增加三角函数积分递推式（I_n记法）
ViperEkura pushed to main at ViperEkura/postgraduate-prep 2026-05-07 20:39:57 +08:00
2a92798bda feat: 增加数学杂项知识（常用公式速查与三角函数口诀）
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 18:05:08 +08:00
6b0a1dbb5e refactor: redesign batching FSM as queue pipeline with dynamic task states
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 14:59:31 +08:00
c05a432e45 chore: use Times New Roman across all scenes, widen Transformer block to 2.4 for \'Transformer Block × 24\' label
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 14:51:33 +08:00
57abefa47f fix: shift GQA layout down 0.4 to avoid title-input overlap
ViperEkura pushed to main at ViperEkura/video-promo 2026-05-07 14:14:17 +08:00
6b26ec33ab chore: switch .gitignore to whitelist mechanism (only .py, .md, .gitignore)