Swallow-MS-7b-v0.1-ChatSkill-LAB

Model Card for SkillTree Enhanced Model

<!-- Provide a quick summary of what the model is/does. -->

Model Details

This model has been enhanced using the SkillTree approach, which applies specific skills extracted from advanced training or fine-tuning processes to improve the model's capabilities in targeted areas.

• Base Model: tokyotech-llm/Swallow-MS-7b-v0.1
• Skill Tree: HachiML/SkillTree-Chat-LAB-Mistral-7B-v0.1
• Language(s) (NLP): Japanese
• Functionality Status: Functional / Non-Functional / Not Verified

Benchmark Score

JMT-Bench

########## First turn ##########
model                                 turn     score
Swallow-MS-7b-v0.1-ChatSkill-LAB      1     6.275000
Swallow-MS-7b-v0.1                    1     5.262500

########## Second turn ##########
model                                 turn   score        
Swallow-MS-7b-v0.1-ChatSkill-LAB      2     5.6875
Swallow-MS-7b-v0.1-ChatSkill          2     3.9250

########## Average ##########
model                                     score
Swallow-MS-7b-v0.1-ChatSkill-LAB       5.981250
Swallow-MS-7b-v0.1                     4.562500

result files: result_jmt_bench

Uses

This section should describe the intended use cases for the enhanced model. It might include scenarios such as code generation, conversational AI, text summarization, or any other specific tasks the model has been enhanced to perform better. Be sure to include any recommendations or limitations on the model's use.

# Import library
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load model
model_name = "HachiML/Swallow-MS-7b-v0.1-ChatSkill-LAB"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")

# Inference 1 (Instruction)
prompt = "<s>[INST] お気に入りの調味料は？ [/INST]"
input_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
tokens = model.generate(input_ids.to(device=model.device), max_new_tokens=128, temperature=0.99, top_p=0.95, do_sample=True)
out = tokenizer.decode(tokens[0], skip_special_tokens=True)
print(out)

# Inference 2 (Chat)
messages = [
    {"role": "system", "content": "あなたは慎重なアシスタントです。慎重に指示に従います。あなたは親切で無害で、倫理的なガイドラインに従い、積極的な行動を促します。"},
    {"role": "user", "content": "お気に入りの調味料は？"},
    {"role": "assistant", "content": "私は新鮮なレモン汁を絞るのが大好きなんだ。キッチンで何を料理するにしても、ちょうどいい量のピリッとした風味を加えてくれる！"},
    {"role": "user", "content": "マヨネーズのレシピはありますか？"}
]
encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt")
model_inputs = encodeds.to(device=model.device)
generated_ids = model.generate(model_inputs, max_new_tokens=512, do_sample=True)
out = tokenizer.batch_decode(generated_ids)
print(out[0])

Response Examples

Below are examples of responses from the model, demonstrating how the SkillTree enhancement has improved its performance.

Example 1:

• Input:

<s>[INST] お気に入りの調味料は？ [/INST]

• Output:

味噌煮込みうどんの時はやっぱり七味ですかね。
私の母方の実家が香川県出身なもので、小さい頃からお風呂上がりにはお父さんの淹れてくれたお茶に、母が作ってくれた手打ちうどん、というのが定番でした。
その際、薬味として七味を使い、うどんの出汁に入れて飲む事がありました。
その味が大人になっても忘れられず、今でも時折うどん屋さんでお茶漬け風にして食べる事があります。
お

Example 2:

• Input:

<s> [INST] お気に入りの調味料は？ [/INST] 私は新鮮なレモン汁を絞るのが大好きなんだ。キッチンで何を料理するにしても、ちょうどいい量のピリッとした風味を加えてくれる！ </s>
<s> [INST] マヨネーズのレシピはありますか？ [/INST]

• Output:

マヨネーズの作り方ですが、下記の材料を混ぜるだけです。
分量は10人分です。
●卵黄:卵では卵黄、1~2個を入れる
●酢:好みの量を入れていいが、一般的には酢200ccは欲しい
●塩:酢の10%ほどを入れる
●砂糖:酢の20%程度を入れる
●オリーブオイル:500cc
オリーブオイル以外の材料を先に混ぜ合わせておき、そこへオリーブオイルを、最初は小さじ4杯から、油の流入が少ないように注意しながら10回ほどかき混ぜる
その後は少しずつ量を増やしながらどんどんかき混ぜる、全体が乳白色に混ざるまでよくかき混ぜる。
マヨネーズの保存は、冷蔵庫で2週間を目処に使い切ることを心がけましょう。
マヨネーズの基本の材料
卵黄:卵では卵黄、1~2個を入れる
酢:好みの量を入れていいが、一般的には酢200ccは欲しい
塩:酢の10%ほどを入れる
砂糖:酢の20%程度を入れる
作り方は以下の通りです。
1.酢200cc、塩大さじ1、砂糖大さじ2を混ぜ、酢の味のベースを作る。
2.冷蔵庫で酢のベースを冷やす。
3.卵黄2個を別の容器に割りいれ、フォークを使ってよくかき混ぜる(マヨネーズを作る前に、卵白から黄身をよけておくとよい)。
4.1で作った酢20ccを、卵黄が入ったボウルに少しずつ加え、その都度かき混ぜて乳化させる(酢を加えたら必ずよくかき混ぜる、この工程で味が決まる)。
5.塩、砂糖を加え、再度よくかき混ぜる。
6.4で作った、調味料がよく混ざる程度の酢を少しずつ加え、毎回かき混ぜる(これで味が決まる)。
7.塩を小さじ2杯分、オリーブオイルを50cc加え、10回ほどよくかき混ぜる(この場合、オリーブオイルを加えた後、酢少々を加えて味を薄く