 Georg Bauer
|
48c2f751b4
|
Align DeepSeek and GLM execution with DS4
|
2026-09-11 17:18:45 +02:00 |
|
|
Georg Bauer
|
02db0968ae
|
Save inference parity implementation and evaluation harness
|
2026-09-10 22:45:59 +02:00 |
|
|
Georg Bauer
|
bd6353804b
|
Optimize Qwen3.8 inference on Apple silicon
|
2026-09-04 23:27:09 +02:00 |
|
|
Georg Bauer
|
bf82df77cb
|
Add native Qwen MTP speculation
|
2026-09-03 22:38:52 +02:00 |
|
|
Georg Bauer
|
714b39f7f3
|
Implement exact Qwen sparse attention
|
2026-09-03 21:45:39 +02:00 |
|
|
Georg Bauer
|
87ccf67d0c
|
Stream Qwen PLE embeddings
|
2026-09-03 21:12:20 +02:00 |
|
|
Georg Bauer
|
c414640050
|
Implement Qwen text core
|
2026-09-03 20:31:53 +02:00 |
|
|
Georg Bauer
|
3773cfda2e
|
Integrate Qwen3.8 model intake
|
2026-09-03 19:56:23 +02:00 |
|
|
Georg Bauer
|
d40e86e5ef
|
correct thinking modes for each model
|
2026-09-01 21:03:13 +02:00 |
|
|
Georg Bauer
|
ee515ee824
|
Load GLM vision weights on demand
|
2026-09-01 20:08:36 +02:00 |
|
|
Georg Bauer
|
46d6a976a5
|
Add GLM 5.3 Flash support
|
2026-09-01 19:06:20 +02:00 |
|
|
Georg Bauer
|
0fa15bb68b
|
Remove deprecated Flash preview and legacy MTP
|
2026-08-31 08:04:20 +02:00 |
|
|
Georg Bauer
|
6c4e792c8a
|
Accelerate DeepSeek DSpark inference
|
2026-08-30 21:45:16 +02:00 |
|
|
Georg Bauer
|
28231e1faf
|
Add end-to-end MXFP4 Metal support
|
2026-08-30 20:42:23 +02:00 |
|
|
Georg Bauer
|
1ac559bbd0
|
Reach DS4 Metal performance parity
|
2026-08-30 15:26:27 +02:00 |
|
|
Georg Bauer
|
f1c177b754
|
Support DeepSeek V4 Flash 0731
|
2026-08-29 20:28:50 +02:00 |
|
|
Georg Bauer
|
d1c380f79d
|
feat: DeepSeek V4 Flash 0731 added
|
2026-08-01 08:40:06 +02:00 |
|
|
Georg Bauer
|
92f351ae0b
|
chore: refactorings for more code reorganizing
|
2026-07-29 19:28:28 +02:00 |
|
|
Georg Bauer
|
9cf1c70a67
|
Keep runtime files inside application support
|
2026-07-28 18:34:08 +02:00 |
|
|
Georg Bauer
|
47b405e72b
|
fix: ssd streaming and deepseek fixes
|
2026-07-26 21:50:59 +02:00 |
|
|
Georg Bauer
|
0d80c217c4
|
Finish DS4 execution parity
|
2026-07-26 20:44:23 +02:00 |
|
|
Georg Bauer
|
4420b81117
|
Integrate DS4 execution parity in Rust
|
2026-07-26 17:58:05 +02:00 |
|
|
Georg Bauer
|
65c9cbfc45
|
Add GLM 5.2 Metal execution
|
2026-07-26 12:16:30 +02:00 |
|
|
Georg Bauer
|
1de954b579
|
Support DeepSeek V4 Pro execution
|
2026-07-26 12:02:06 +02:00 |
|
|
Georg Bauer
|
3c75b8f6c1
|
Support long-running agent turns
|
2026-07-26 11:07:28 +02:00 |
|
|
Georg Bauer
|
2a14b93335
|
Make compaction transitions recoverable
|
2026-07-26 10:56:19 +02:00 |
|
|
Georg BauerandClaude Opus 5
|
1f27270d66
|
Reclaim KV checkpoints the index cannot reach
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
2026-07-26 09:25:37 +02:00 |
|
|
Georg BauerandClaude Opus 5
|
63e2a74ad5
|
Manage the KV cache disc usage from preferences and stats
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
2026-07-26 09:10:18 +02:00 |
|
|
Georg Bauer
|
4a45735072
|
Add persistent local agent tools
|
2026-07-25 14:57:17 +02:00 |
|
|
Georg Bauer
|
a1761fa731
|
Add prefix-aware KV session cache
|
2026-07-25 13:56:10 +02:00 |
|
|
Georg Bauer
|
292ee82af5
|
Make tokenizer byte offsets total
|
2026-07-25 13:09:04 +02:00 |
|
|
Georg Bauer
|
3708afb1b6
|
Document Metal executor drop order
|
2026-07-25 13:04:09 +02:00 |
|
|
Georg Bauer
|
44fce6585c
|
Document Metal buffer view ownership
|
2026-07-25 13:02:13 +02:00 |
|
|
Georg Bauer
|
8f91a06336
|
Recover the model runtime from panics
|
2026-07-25 13:00:20 +02:00 |
|
|
Georg Bauer
|
4292e0d3f9
|
Split Rust code into domain modules
|
2026-07-25 11:22:59 +02:00 |
|
|
Georg Bauer
|
2fbe605b5d
|
Implement batched server parity and observability
|
2026-07-25 08:32:35 +02:00 |
|
|
Georg Bauer
|
3154f57a2f
|
full chat implemented
|
2026-07-24 19:59:41 +02:00 |
|
|
Georg Bauer
|
36946e6e5f
|
Complete long-context chat support
|
2026-07-24 18:16:34 +02:00 |
|
|
Georg Bauer
|
5ab110ed0b
|
feat: better chat support for thinking
|
2026-07-24 18:01:53 +02:00 |
|
|
Georg Bauer
|
ff984bb96a
|
feat: first cut at actual token generation and model loading
|
2026-07-24 17:44:41 +02:00 |
|
|
Georg Bauer
|
616b550849
|
Implement DS4 model intake
|
2026-07-24 16:53:03 +02:00 |
|