Megac4ai-command-r-plus

Megac4ai-command-r-plus

GGUF version is here.

🚨 This model is created using the special mergekit that supports c4ai-command-r-plus.

This is a 160b frankenmerge model created by interleaving layers of CohereForAI/c4ai-command-r-plus with itself using mergekit.

Output comparison

Test Case Details

Condition: temperature=0.3

<|START_OF_TURN_TOKEN|><|USER_TOKEN|>ティム: やあ、調子はどう？
キム: いろいろやろうとしてたんだけど、また先延ばしにしちゃったよ。
ティム: 何をしようとしていたの？
キム: 大学の課題だよ。どうにもやる気が出なくてね。
ティム: 集中できないなら、ポモドーロ・テクニックをするといいよ。
キム: 何それ？
ティム: 25分作業して、5分休憩するのを繰り返すんだよ。一回あたりの作業時間が短くて集中できるよ。
キム: うーん、集中っていうわけじゃないんだよね
ティム: じゃあ1日に5分だけでいいから机で課題をするっていうのはどう？
キム: 5分じゃ何もできなくない？
ティム: 短い時間でもいいから机で作業するっていうのがポイントなんだよ。むしろもっとやりたい、くらいで止めておくと毎日続くもっと長い時間できるようになるよ。
キム: 確かにそれならできるかも。ありがとう！
Q: キムは何をやってみようとしていますか？また何故それをやろうとしていますか？<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>

This task is included in elyza/ELYZA-tasks-100.

Output Example

Test environment

This model was tested using text-generation-webui. I use preset min_p and Null preset with temperature=0.3 for Generation.

Usage

Please install transformers from the source repository that includes the necessary changes for this model.

# pip install 'git+https://github.com/huggingface/transformers.git'
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "nitky/megac4ai-command-r-plus"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

# Format message with the command-r-plus chat template
messages = [{"role": "user", "content": "Hello, how are you?"}]
input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>

gen_tokens = model.generate(
    input_ids, 
    max_new_tokens=100, 
    do_sample=True, 
    temperature=0.3,
    )

gen_text = tokenizer.decode(gen_tokens[0])
print(gen_text)

Quantized model through bitsandbytes, 4-bit precision

# pip install 'git+https://github.com/huggingface/transformers.git' bitsandbytes accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(load_in_4bit=True)

model_id = "nitky/megac4ai-command-r-plus"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config)

# Format message with the command-r-plus chat template
messages = [{"role": "user", "content": "Hello, how are you?"}]
input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>

gen_tokens = model.generate(
    input_ids, 
    max_new_tokens=100, 
    do_sample=True, 
    temperature=0.3,
    )

gen_text = tokenizer.decode(gen_tokens[0])
print(gen_text)

Merge Details

Merge Method

This model was merged using the passthrough merge method.

Models Merged

The following models were included in the merge:

• CohereForAI/c4ai-command-r-plus

Configuration

The following YAML configuration was used to produce this model:

dtype: float16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 20]
    model: CohereForAI/c4ai-command-r-plus
- sources:
  - layer_range: [11, 31]
    model: CohereForAI/c4ai-command-r-plus
- sources:
  - layer_range: [22, 42]
    model: CohereForAI/c4ai-command-r-plus
- sources:
  - layer_range: [33, 53]
    model: CohereForAI/c4ai-command-r-plus
- sources:
  - layer_range: [44, 64]
    model: CohereForAI/c4ai-command-r-plus