Make videos on the computer you already own. FreeVideo runs MiniMax H3 in as little as 8 GB of VRAM and 16 GB of RAM, and adapts its acceleration path to your hardware.
docs: update community results [skip ci] (#173)
Bowen Xue8aa888e894eb
diff --git a/docs/community-results.md b/docs/community-results.mdindex 25434b1..c4f014e 100644--- a/docs/community-results.md+++ b/docs/community-results.md@@ -6,7 +6,7 @@ Thanks to the community members who sent diagnostic reports. FreeVideo already r|---|---:|---:|---:|---:|:---:|:---:|---:|:---:|---:|| 4060 Ti | 16 GB | 32 GB | 1920×1088 | 15 s | Light | 0.3.0 | 3306 s | 0.3.6 | 2730 s (−580 s) || 5060 | 8 GB | 16 GB | 1344×768 | 15 s | Light | 0.3.1 | 1367 s | 0.3.6 | 1160 s (−210 s) |-| 3070 Ti | 8 GB | 96 GB | 1344×768 | 10 s | Light | 0.3.1 | 749 s | 0.3.6 | 690 s (−60 s) |+| 3070 Ti | 8 GB | 96 GB | 1344×768 | 10 s | Light | 0.3.1 | 749 s | 0.3.22 | 665 s (−80 s)⁶ || 4060 | 8 GB | 32 GB | 768×1344 | 15 s | Light | 0.3.4 | 1592 s | 0.3.10 | 1450 s (−140 s) || 5070 Laptop | 8 GB | 32 GB | 1344×768 | 5 s | Light | 0.3.4 | 555 s | 0.3.10 | 485 s (−70 s) || PRO 5000 | 72 GB | 128 GB | 768×1344 | 10 s | High | 0.3.4 | 552 s | 0.3.4 | Already optimal |@@ -29,6 +29,16 @@ Thanks to the community members who sent diagnostic reports. FreeVideo already r| 5070 Ti Laptop | 12 GB | 32 GB | 1344×768 | 10 s | High | 0.3.20 | 1545 s | 0.3.21 | 1310 s (−240 s) || 5060 Ti | 16 GB | 64 GB | 1344×768 | 10 s | Light | 0.3.20 | 471 s | 0.3.21 | 465 s (−10 s) || 4070 SUPER | 12 GB | 32 GB | 1344×768 | 15 s | High | 0.3.20 | 2966 s | 0.3.21 | 2050 s (−920 s) |+| 4070 Laptop | 8 GB | 32 GB | 544×960 | 3 s | Light⁵ | 0.3.22 | 133 s | 0.3.22 | Already optimal |+| M4 Pro | Unified | 24 GB | 960×544 | 10 s | Light | 0.3.8 | 3457 s | 0.3.8 | Already optimal |+| 5090 | 32 GB | 64 GB | 1376×768 | 10 s | Light | 0.3.20 | 294 s | 0.3.23 | 265 s (−30 s) |+| 5060 Laptop | 8 GB | 16 GB | 960×544 | 3 s | Light | 0.3.21 | 908 s | 0.3.23 | 820 s (−90 s) |+| 3060 | 12 GB | 64 GB | 1184×672 | 10 s | Light⁵ | 0.3.22 | 604 s | 0.3.23 | 565 s (−40 s) |+| 4070 | 12 GB | 32 GB | 960×544 | 10 s | Light | 0.3.22 | 340 s | 0.3.22 | Already optimal |+| 5090 | 32 GB | 64 GB | 1344×768 | 10.8 s | Light | 0.3.22 | 249 s | 0.3.22 | Already optimal |+| 3090 | 24 GB | 64 GB | 1344×768 | 10 s | Light | 0.3.23 | 704 s | 0.3.23 | Already optimal |+| 5060 Ti | 16 GB | 32 GB | 1344×768 | 12.3 s | Light | 0.3.23 | 573 s | 0.3.23 | Already optimal |+| 5080 | 16 GB | 64 GB | 1344×768 | 10 s | Light | 0.3.23 | 240 s | 0.3.23 | Already optimal |- Time: end to end, from submission until the video is saved.- After: the estimated time, derived from the report's stage times and from complete planned runs on our test machines (RTX 3090 and RTX 5060 Ti on Windows, RTX PRO 6000 on Linux) under the reporting machine's VRAM and RAM limits, not from reruns on that machine.@@ -37,3 +47,5 @@ Thanks to the community members who sent diagnostic reports. FreeVideo already r- ² This machine still runs the FP8 model; the estimate is for the int8 model, a one-click upgrade in the launcher since 0.3.0.- ³ Preview: only the first pass, at half resolution. With 16 GB of RAM most of the model is read from the drive at every step (about 80 MB/s on this machine), which takes most of the time.- ⁴ The report had no diagnostic file; the estimate replays the plan for this machine's configuration and measures the same request on our test machines.+- ⁵ Second pass after a preview: the time covers the latent upscale, the three second-pass steps and decoding, not the first pass, which made the preview in an earlier request.+- ⁶ Measured on the same machine with 0.3.22.diff --git a/docs/zh-CN/community-results.md b/docs/zh-CN/community-results.mdindex 5c79f61..a648d0a 100644--- a/docs/zh-CN/community-results.md+++ b/docs/zh-CN/community-results.md@@ -6,7 +6,7 @@|---|---:|---:|---:|---:|:---:|:---:|---:|:---:|---:|| 4060 Ti | 16 GB | 32 GB | 1920×1088 | 15 s | 轻量 | 0.3.0 | 3306 s | 0.3.6 | 2730 s(−580 s) || 5060 | 8 GB | 16 GB | 1344×768 | 15 s | 轻量 | 0.3.1 | 1367 s | 0.3.6 | 1160 s(−210 s) |-| 3070 Ti | 8 GB | 96 GB | 1344×768 | 10 s | 轻量 | 0.3.1 | 749 s | 0.3.6 | 690 s(−60 s) |+| 3070 Ti | 8 GB | 96 GB | 1344×768 | 10 s | 轻量 | 0.3.1 | 749 s | 0.3.22 | 665 s(−80 s)⁶ || 4060 | 8 GB | 32 GB | 768×1344 | 15 s | 轻量 | 0.3.4 | 1592 s | 0.3.10 | 1450 s(−140 s) || 5070 笔记本 | 8 GB | 32 GB | 1344×768 | 5 s | 轻量 | 0.3.4 | 555 s | 0.3.10 | 485 s(−70 s) || PRO 5000 | 72 GB | 128 GB | 768×1344 | 10 s | 精细 | 0.3.4 | 552 s | 0.3.4 | 已是最优 |@@ -29,6 +29,16 @@| 5070 Ti 笔记本 | 12 GB | 32 GB | 1344×768 | 10 s | 精细 | 0.3.20 | 1545 s | 0.3.21 | 1310 s(−240 s) || 5060 Ti | 16 GB | 64 GB | 1344×768 | 10 s | 轻量 | 0.3.20 | 471 s | 0.3.21 | 465 s(−10 s) || 4070 SUPER | 12 GB | 32 GB | 1344×768 | 15 s | 精细 | 0.3.20 | 2966 s | 0.3.21 | 2050 s(−920 s) |+| 4070 笔记本 | 8 GB | 32 GB | 544×960 | 3 s | 轻量⁵ | 0.3.22 | 133 s | 0.3.22 | 已是最优 |+| M4 Pro | 统一内存 | 24 GB | 960×544 | 10 s | 轻量 | 0.3.8 | 3457 s | 0.3.8 | 已是最优 |+| 5090 | 32 GB | 64 GB | 1376×768 | 10 s | 轻量 | 0.3.20 | 294 s | 0.3.23 | 265 s(−30 s) |+| 5060 笔记本 | 8 GB | 16 GB | 960×544 | 3 s | 轻量 | 0.3.21 | 908 s | 0.3.23 | 820 s(−90 s) |+| 3060 | 12 GB | 64 GB | 1184×672 | 10 s | 轻量⁵ | 0.3.22 | 604 s | 0.3.23 | 565 s(−40 s) |+| 4070 | 12 GB | 32 GB | 960×544 | 10 s | 轻量 | 0.3.22 | 340 s | 0.3.22 | 已是最优 |+| 5090 | 32 GB | 64 GB | 1344×768 | 10.8 s | 轻量 | 0.3.22 | 249 s | 0.3.22 | 已是最优 |+| 3090 | 24 GB | 64 GB | 1344×768 | 10 s | 轻量 | 0.3.23 | 704 s | 0.3.23 | 已是最优 |+| 5060 Ti | 16 GB | 32 GB | 1344×768 | 12.3 s | 轻量 | 0.3.23 | 573 s | 0.3.23 | 已是最优 |+| 5080 | 16 GB | 64 GB | 1344×768 | 10 s | 轻量 | 0.3.23 | 240 s | 0.3.23 | 已是最优 |- 用时:从提交到视频保存完成的端到端用时。- 优化后:预估用时,根据报告中的分段用时,结合测试机(Windows RTX 3090 和 RTX 5060 Ti、Linux RTX PRO 6000)在报告机器的显存和内存上限下运行完整规划的实测推算,不是在报告机器上重跑。@@ -37,3 +47,5 @@- ² 这台机器仍在使用 FP8 模型;预估为在启动器中一键升级到 int8 模型(0.3.0 起提供)后的用时。- ³ 预览:只运行第一遍(半分辨率)。16 GB 内存放不下大部分模型,每一步都要从硬盘读取(这台机器约 80 MB/s),大部分时间花在读盘上。- ⁴ 报告未附诊断文件;预估按这台机器的配置重放规划,并在测试机上实测同样的请求。+- ⁵ 预览之后的第二遍:用时包括潜变量放大、第二遍的 3 步和解码,不含之前那次请求里生成预览的第一遍。+- ⁶ 同一台机器使用 0.3.22 的实测用时。