diff options
| -rw-r--r-- | .github/workflows/book-cd.yml | 2 | ||||
| -rw-r--r-- | candle-book/src/inference/hub.md | 4 | ||||
| -rw-r--r-- | candle-examples/src/lib.rs | 8 |
3 files changed, 6 insertions, 8 deletions
diff --git a/.github/workflows/book-cd.yml b/.github/workflows/book-cd.yml index fc693a78..e8149e38 100644 --- a/.github/workflows/book-cd.yml +++ b/.github/workflows/book-cd.yml @@ -1,7 +1,5 @@ name: Deploy Rust book on: - # TODO put this back only when merging after this PR lands. - pull_request: push: branches: - main diff --git a/candle-book/src/inference/hub.md b/candle-book/src/inference/hub.md index a974a1fa..b924b76d 100644 --- a/candle-book/src/inference/hub.md +++ b/candle-book/src/inference/hub.md @@ -67,8 +67,8 @@ let bias = weights.get("bert.encoder.layer.0.attention.self.query.bias").unwrap( let linear = Linear::new(weight.clone(), Some(bias.clone())); -let input_ids = Tensor::zeros((3, 7680), DType::F32, &Device::Cpu).unwrap(); -let output = linear.forward(&input_ids); +let input_ids = Tensor::zeros((3, 768), DType::F32, &Device::Cpu).unwrap(); +let output = linear.forward(&input_ids).unwrap(); ``` For a full reference, you can check out the full [bert](https://github.com/LaurentMazare/candle/tree/main/candle-examples/examples/bert) example. diff --git a/candle-examples/src/lib.rs b/candle-examples/src/lib.rs index 3410026e..2b6009b4 100644 --- a/candle-examples/src/lib.rs +++ b/candle-examples/src/lib.rs @@ -73,8 +73,8 @@ let mmap = unsafe { Mmap::map(&file).unwrap() }; // Use safetensors directly let tensors = SafeTensors::deserialize(&mmap[..]).unwrap(); let view = tensors -.tensor("bert.encoder.layer.0.attention.self.query.weight") -.unwrap(); + .tensor("bert.encoder.layer.0.attention.self.query.weight") + .unwrap(); // We're going to load shard with rank 1, within a world_size of 4 // We're going to split along dimension 0 doing VIEW[start..stop, :] @@ -86,7 +86,7 @@ let mut tp_shape = view.shape().to_vec(); let size = tp_shape[0]; if size % world_size != 0 { -panic!("The dimension is not divisble by `world_size`"); + panic!("The dimension is not divisble by `world_size`"); } let block_size = size / world_size; let start = rank * block_size; @@ -102,7 +102,7 @@ tp_shape[dim] = block_size; // Convert safetensors Dtype to candle DType let dtype: DType = dtype.try_into().unwrap(); -// TODO: Implement from_buffer_iterator to we can skip the extra CPU alloc. +// TODO: Implement from_buffer_iterator so we can skip the extra CPU alloc. let raw: Vec<u8> = iterator.into_iter().flatten().cloned().collect(); let tp_tensor = Tensor::from_raw_buffer(&raw, dtype, &tp_shape, &Device::Cpu).unwrap(); // ANCHOR_END: book_hub_3 |