Raunak Shah

I have loaded an LLM in huggingface with `load_in_8bit=True`. I noticed the objects in the state_dict are structured something like 1. `model.layers.18.self_attn.k_proj.weight` 2. `model.layers.18.self_attn.k_proj.SCB` 3. `model.layers.18.self_attn.k_proj.weight_format` The `SCB` and `weight_format`...

Quanto scale values seem unpopulated in quantized model

2

When loading a mistral model I noticed that the `output_scale` and `input_scale` values associated with the quantized tensors were just tensors with the value 1, i.e. `tensor(1., device='cuda:0')` This seems...

Changes in Acto for running tests on kubeblocks postgresql operator

2

Mainly 3 changes - 1. Added scope for running `kubectl create -f file` along with apply and wait commands. 2. Removed server-side flag while running `kubectl apply` since that prevented...

feat: update merge_insert to add statistics for inserted, updated, deleted rows

2

Addresses https://github.com/lancedb/lance/issues/2019

Many things to add/change, but wanted to get a review on the initial structure. Some considerations - 1. Can add write benchmarking within the same functions, and populate results in...

feat: add a packed struct encoding to lance

1

Introduces a new `PackedStruct` encoding, should speed up random access for struct data, ref #2601 - Can currently support non-nullable, primitive fixed-length types (including fixed size list) - Implemented as...

Currently struct arrays are encoded using multiple IOPS for random access, which is suboptimal. e.g. if we have a struct array `x: {1, 2, 3}, y: {2.4, 5.6, 3.8}, z:...

feat: added edge case tests for v2 writer

1

1. Writes with zero batches 2. Writes with zero rows (all batches) 3. Writes with zero rows (some batches) Not sure whether we should validate the data after writing the...

Currently all tests for binary arrays end up generating binary arrays where all items have the same size (see `RandomBinaryGenerator::generate()` in `generator.rs`) We should update this, (i) for the sake...

Currently the base string encoding in the encoding tree is a binary encoding. This uses 2 IOPS, one for encoding the offsets, and one for encoding the bytes. However we...