Nano2527 commited on
Commit
e29db74
·
verified ·
1 Parent(s): 05705a6

Upload 8 files

Browse files
README.md CHANGED
@@ -1,3 +1,24 @@
1
  ---
2
- license: mit
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: apache-2.0
3
+ library_name: transformers
4
+ pipeline_tag: text-generation
5
+ tags:
6
+ - datathai
7
+ - thai
8
+ - toy-model
9
+ - safetensors
10
  ---
11
+
12
+ # DataThai-1M-playground-improved
13
+
14
+ Playable tiny Thai-English toy model for pipeline testing.
15
+
16
+ This is **not** DataThai-1B. It is a ~1M parameter smoke/playground model trained on a tiny QA-style demo corpus.
17
+
18
+ Use it for:
19
+ - loading test
20
+ - generation test
21
+ - Hugging Face deploy test
22
+ - sharing a small runnable model artifact
23
+
24
+ Do not use it as a real language model benchmark.
artifact_summary.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "artifact_dir": "/content/datathai-deploy/DataThai-QwenStyle-smoke-qwen2",
3
+ "repo_id": "SPAISS6F1/DataThai-QwenStyle-smoke",
4
+ "size": "smoke",
5
+ "arch": "qwen2",
6
+ "parameters": 902144,
7
+ "status": "tiny smoke-trained",
8
+ "safetensors": [
9
+ "model.safetensors"
10
+ ],
11
+ "index_files": [],
12
+ "name": "DataThai-1M-playground-improved",
13
+ "note": "Improved tiny playable smoke model, not a real 1B LLM.",
14
+ "recommended_repo_name": "DataThai-1M-playground-improved"
15
+ }
config.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen2ForCausalLM"
4
+ ],
5
+ "attention_dropout": 0.0,
6
+ "bos_token_id": 0,
7
+ "dtype": "float32",
8
+ "eos_token_id": 0,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 128,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 384,
13
+ "layer_types": [
14
+ "full_attention",
15
+ "full_attention"
16
+ ],
17
+ "max_position_embeddings": 512,
18
+ "max_window_layers": 28,
19
+ "model_type": "qwen2",
20
+ "num_attention_heads": 4,
21
+ "num_hidden_layers": 2,
22
+ "num_key_value_heads": 1,
23
+ "pad_token_id": 6,
24
+ "rms_norm_eps": 1e-06,
25
+ "rope_parameters": {
26
+ "rope_theta": 1000000.0,
27
+ "rope_type": "default"
28
+ },
29
+ "sliding_window": null,
30
+ "tie_word_embeddings": true,
31
+ "transformers_version": "5.10.2",
32
+ "use_cache": true,
33
+ "use_sliding_window": false,
34
+ "vocab_size": 4096
35
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "eos_token_id": 0,
5
+ "output_attentions": false,
6
+ "output_hidden_states": false,
7
+ "pad_token_id": 6,
8
+ "transformers_version": "5.10.2",
9
+ "use_cache": true
10
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4514ff9e19fa0891f0a7b76f23daae2ba3ba7a2e7b1c9d235a1fdd9b3ba82aeb
3
+ size 3611320
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": null,
3
+ "backend": "tokenizers",
4
+ "bos_token": "<|endoftext|>",
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|endoftext|>",
7
+ "extra_special_tokens": [
8
+ "<|im_start|>",
9
+ "<|im_end|>",
10
+ "<|system|>",
11
+ "<|user|>",
12
+ "<|assistant|>"
13
+ ],
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 1000000000000000019884624838656,
17
+ "pad_token": "<|pad|>",
18
+ "tokenizer_class": "Qwen2Tokenizer",
19
+ "unk_token": "<|endoftext|>"
20
+ }
tokenizer_seed_corpus.txt ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ ภาษาไทยเป็นภาษาที่สวยงามและมีเอกลักษณ์ DataThai-1B is a Thai-first English-capable LLM.
2
+ การสร้างโมเดลภาษา from scratch ต้องมี tokenizer, data cleaning, pretraining, evaluation, and safety tuning.
3
+ Transformers use attention, RoPE, RMSNorm, and SwiGLU to model long-range dependencies.
4
+ ตัวอย่างโจทย์: ถ้า x = 3 และ y = 5 แล้ว x*y + 2 เท่ากับ 17
5
+ def add(a, b): return a + b # ฟังก์ชันบวกเลขสองจำนวน