-
Currently for N bits in the key as path part, we apply the constraint: `resident * 2^N + path`
The path part is constrainted to less than 2^N, while we have no constraint on `resident` so malice pr…
-
Hi, this model seems nice, but I do find that the inference speed is very slow (70ms/token on single A100), so I want to speed up it.
It seems to be related with MPT itself: https://huggingface.co/…
-
Unexpected behaviour: For models `ggml-mpt-7b-base`, `ggml-mpt-7b-instruct`, `ggml-gpt4all-j-v1`, after the first prompt the model is not downloaded, instead an error occurs `Error: Model filename not…
-
After running this command ```python src/finetune.py --base_model 'mosaicml/mpt-7b-instruct' --data_path 'yahma/alpaca-cleaned' --output_dir './lora-mpt' --lora_target_modules '[Wqkv]'```, i got this …
-
Hi, is the 3b model being slower expected?
-
The current recommendation is to run RP with somewhere betweeen 1-2GiB per shard because there is a certain amount of "static" memory overhead. We don't currently have a good number for what the stati…
-
Merkle (Patricia) tree proofs essentially prove that a certain element (or several elements) belongs to a certain data collection (list in the case of Merkle trees; map in the case of MPTs). Essential…
-
When I install the `anas-awadalla/mpt-1b-redpajama-200b` language encoder from HuggingFace, I get the following warning message:
```
A new version of the following files was downloaded from https:…
-
Add `/sc_version` and `/protocol_version` nodes to MPT to store the current running smart contract and protocol versions. Only smart contract owner could create transactions to update them.
-
This page should have all the partner logos, but laid out fully in rows instead of a carousel (as done on the homepage):
![image](https://uploads.linear.app/399df041-a1eb-4d3d-bb80-9564252bf082/656f9…