Rain-Bus
ffd2defdfc
add Chinese annotations to all source files for learning purposes
...
Annotated 16 source files covering the full architecture:
engine (scheduler, block manager, model runner), layers (attention,
linear, sampler, etc.), model (qwen3), and utils.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com >
2026-05-25 21:33:15 +08:00
9fa256a56d
fix cache hit
2026-04-26 03:49:14 +08:00
f64d821c20
fix chunked prefill bugs and refactor
2026-04-26 02:53:06 +08:00
44a51afc8a
Merge pull request #207 from DestineG/fix-prefill-index-out-of-range
...
fix(scheduler): recalculate num_tokens after allocate to prevent IndexError at model_runner:155
2026-04-25 18:12:43 +08:00
25794a1f29
fix(model_runner): correct seqlen_k to chunk boundary in prepare_prefill
...
During chunked prefill, seqlen_k was set to len(seq) (the full sequence
length), causing the attention kernel to access uninitialized KV slots
for tokens not yet scheduled in the current chunk.
Fix: reorder so that end = start + seqlen_q is computed first, then
set seqlen_k = end — limiting attention to the current chunk boundary.
Fixes #212
2026-04-22 15:13:19 -07:00
77dd709ca1
fix(scheduler): recalculate num_tokens after allocate to prevent IndexError
...
The scheduler overestimated num_scheduled_tokens because it used an outdated num_cached_tokens before block_manager.allocate(seq) could update it via prefix cache hits. In prepare_prefill (model_runner.py), this caused 'end = start + seqlen_q' to exceed the sequence length, leading to an inflated 'end_block'. Consequently, an 'index out of range' error occurred at line 155 when accessing seq.block_table[i] beyond its actual physical allocation.
2026-04-20 16:34:27 +08:00
8d63a98c03
support chunked prefill and fix minor bug
2026-04-14 03:05:35 +08:00
9e8507ef41
minor simplify
2026-04-13 22:09:46 +08:00
02a95fdc66
Merge pull request #203 from Anai-Guo/fix-row-parallel-bias-crash
...
fix RowParallelLinear weight_loader crash when bias is enabled
2026-04-13 21:26:07 +08:00
a4f94cb38b
Merge pull request #200 from KinglittleQ/fix-scheduler-typing
...
Fix scheduler.postprocess return type
2026-04-13 21:13:37 +08:00
00eea73176
Merge pull request #172 from IceCreamMilkyTea/main
...
enable 'slots=True' for dataclasses
2026-04-13 20:45:50 +08:00
52d2215911
Merge pull request #148 from guodongxiaren/main
...
remove hard code for block_size
2026-04-13 20:36:32 +08:00
bf99453d90
fix RowParallelLinear weight_loader crash when bias is enabled
...
When RowParallelLinear has bias=True, the weight_loader crashes with an
IndexError because it calls param_data.size(tp_dim) where tp_dim=1, but
the bias tensor is 1D and only has dimension 0.
The bias in RowParallelLinear is not sharded (all ranks hold the full
bias, only rank 0 applies it), so skip the sharding logic for 1D params.
Fixes GeeeekExplorer/nano-vllm#125
2026-04-12 23:10:52 -07:00
498f5a1aa8
Fix scheduler.postprocess return type
2026-04-11 13:23:49 +08:00
f438ce463f
enable slots=True for dataclasses
2026-02-08 23:46:01 -05:00
55c64e7fdf
remove hard code for block_size
2025-12-30 01:55:17 +08:00
82f5ca244f
fix bug for tp
2025-12-18 01:28:25 +08:00
2f21442653
support qwen2
2025-11-04 01:44:42 +08:00
6ef2a4f630
compile random sampling
2025-08-31 22:55:34 +08:00
df99418f7d
simplify
2025-08-31 20:02:51 +08:00
f5b4840276
fix(model_runner): correct position indexing to be 0-based
...
- Change position calculation from len(seq) to len(seq) - 1
2025-07-04 14:29:12 +08:00
38baf0bbe4
remove assert shape
2025-06-27 23:00:30 +08:00
cb0b3dec3f
remove rng state
2025-06-27 22:50:33 +08:00
1caeec8dfa
same as vllm
2025-06-27 18:50:56 +08:00
658520b788
warmup and allocate
2025-06-27 01:51:57 +08:00
054aec852d
Fix: Division-by-Zero Risk and Typo
2025-06-24 02:02:33 +08:00
03cfc13bb3
faster pickle
2025-06-23 00:51:52 +08:00
cde3fc22c2
simplify
2025-06-21 17:19:15 +08:00
ffafaeb133
Release CUDA Graphs resource before exit.
2025-06-18 16:17:31 +08:00
4fc764f175
Merge pull request #22 from cheunglei/use_spawn
2025-06-17 23:53:59 +08:00
b5ace32982
use spawn
2025-06-17 23:49:15 +08:00
bc0ad5a116
better
2025-06-17 23:33:38 +08:00
7e42fa6f63
fix
2025-06-15 13:28:29 +08:00
326b121fad
Merge pull request #10 from MARD1NO/refine_return_hint_in_schedule
2025-06-15 10:39:51 +08:00
fc778a4da9
better
2025-06-15 10:36:45 +08:00
98bbbefb68
schedule return bool args
2025-06-15 10:15:05 +08:00
53b3ef2e32
support tensor parallel
2025-06-15 01:31:24 +08:00
b6136383c9
support fast pickle
2025-06-14 13:36:57 +08:00
4a8aa090a7
fix
2025-06-14 00:56:07 +08:00
59aa3ff57c
better
2025-06-13 13:07:33 +08:00
98a1551a7d
support CUDA_VISIBLE_DEVICES
2025-06-12 23:14:01 +08:00
ec3c60d96f
update bench
2025-06-12 22:54:51 +08:00
f16adb729e
refactor
2025-06-12 09:41:12 +08:00
fee58d44e4
fix
2025-06-12 01:00:31 +08:00
08c84ec08d
multi file loader
2025-06-12 01:00:09 +08:00
386290d69e
refactor
2025-06-11 21:12:57 +08:00
b98e1ca305
fix
2025-06-10 21:25:54 +08:00
a5a4909e6a
init commit
2025-06-10 00:27:01 +08:00
Rain-Bus
GeekExplorer