You can, but it would be very very slow. You really want a GPU.
The training code for "v2" will be on the repo soon, and you could use that to train from a smaller Pythia model.
Maybe the team will just do that. But models small enough to work on CPUs are <100M params and that may not perform that well for the kind of text-gen QA task people expect to use this for.

Can we quantize the model, like someone did in Llama.cpp? (Pardon my ignorance)

Sure, you can try. See https://github.com/databrickslabs/dolly for source code (2.0 training code coming soon)

People have already done it here: https://github.com/ggerganov/llama.cpp/discussions/569. Looks like it runs on CPU just fine :)

Direct link to cpu ready version is here - https://huggingface.co/geemili/dolly-v2-12b/tree/main

How do you run https://huggingface.co/geemili/dolly-v2-12b/tree/main (12b-quantized model, .ggml file) with AutoTokenizer and AutoModelForCausalLM?

How do you run https://huggingface.co/geemili/dolly-v2-12b/tree/main (12b-quantized model, .ggml file) with AutoTokenizer and AutoModelForCausalLM?

You can try the following in the attachment.

How do you run https://huggingface.co/geemili/dolly-v2-12b/tree/main (12b-quantized model, .ggml file) with AutoTokenizer and AutoModelForCausalLM?

You can try the following in the attachment.

Having trouble installing bitsnadbytes on databricks E8as_v4 GPU cluster?

Hm, what issue? bitsandbytes has been working fine for me

Library installation attempted on the driver node of cluster 0413-233703-4jtufovq and failed. Please refer to the following error message to fix the library or contact Databricks support. Error Code: DRIVER_LIBRARY_INSTALLATION_FAILURE. Error Message: org.apache.spark.SparkException: Process List(bash, /local_disk0/.ephemeral_nfs/cluster_libraries/python/python_start_clusterwide.sh, /local_disk0/.ephemeral_nfs/cluster_libraries/python/bin/pip, install, bitsandbytes==0.38.1, --index-url, https://github.com/timdettmers/bitsandbytes, --disable-pip-version-check) exited with code 1. ERROR: Could not find a version that satisfies the requirement bitsandbytes==0.38.1 (from versions: none) ERROR: No matching distribution found for bitsandbytes==0.38.1

Works fine for me: %pip install bitsandbytes==0.38.1 using the 13.0 ML runtime. How are you installing, exactly?

using %pip install bitsandbytes==0.38.1

Welcome to bitsandbytes. For bug reports, please run

python -m bitsandbytes

and submit this information together with your error trace to: https://github.com/TimDettmers/bitsandbytes/issues

bin /local_disk0/.ephemeral_nfs/envs/pythonEnv-6fa3d848-9028-4c25-be76-e27f73042d8f/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cuda117_nocublaslt.so
CUDA_SETUP: WARNING! libcudart.so not found in any environmental path. Searching in backup paths...
CUDA SETUP: CUDA runtime path found: /usr/local/cuda/lib64/libcudart.so
CUDA SETUP: Highest compute capability among GPUs detected: 7.0
CUDA SETUP: Detected CUDA version 117
CUDA SETUP: Loading binary /local_disk0/.ephemeral_nfs/envs/pythonEnv-6fa3d848-9028-4c25-be76-e27f73042d8f/lib/

/local_disk0/.ephemeral_nfs/envs/pythonEnv-6fa3d848-9028-4c25-be76-e27f73042d8f/lib/python3.10/site-packages/bitsandbytes/cuda_setup/main.py:145: UserWarning: WARNING: The following directories listed in your path were found to be non-existent: {PosixPath('/databricks/jars/*')}
warn(msg)
/local_disk0/.ephemeral_nfs/envs/pythonEnv-6fa3d848-9028-4c25-be76-e27f73042d8f/lib/python3.10/site-packages/bitsandbytes/cuda_setup/main.py:145: UserWarning: Found duplicate ['libcudart.so', 'libcudart.so.11.0', 'libcudart.so.12.0'] files: {PosixPath('/usr/local/cuda/lib64/libcudart.so'), PosixPath('/usr/local/cuda/lib64/libcudart.so.11.0')}.. We'll flip a coin and try one of these, in order to fail forward.
Either way, this might cause trouble in the future:
If you get CUDA error: invalid device function errors, the above might be the cause and the solution is to make sure only one ['libcudart.so', 'libcudart.so.11.0', 'libcudart.so.12.0'] in the paths that we search based on your env.
warn(msg)
/local_disk0/.ephemeral_nfs/envs/pythonEnv-6fa3d848-9028-4c25-be76-e27f73042d8f/lib/python3.10/site-packages/bitsandbytes/cuda_setup/main.py:145: UserWarning: WARNING: Compute capability < 7.5 detected! Only slow 8-bit matmul is supported for your GPU!
warn(msg)

Then when I run print(pipe("Explain to me the difference between nuclear fission and fusion."))

RuntimeError: probability tensor contains either inf, nan or element < 0

Yeah, that's all working - it's just that the model hits overflow on that input in 8-bit. This can happen. IIRC this seemed to happen on the V100, not A10, but may be just coincidence. Try an A10, or a smaller model.