Hugging Face's logo

Datasets:
jed351
/
Nemotron-3-Nano-RL-Training-Blend-STEM-Yue-Translated

Languages:
Yue Chinese
Chinese
Dataset card Data Studio Files
xet
 Community
jed351 commited on
Commit
d48966d
·
verified ·
1 Parent(s): 8610876

Upload folder using huggingface_hub

Browse files
Files changed (5) hide show
  1. .gitattributes +3 -0
  2. README.md +14 -4
  3. en/nano_v3_sft_profiled_comp_coding_50tests.jsonl +3 -0
  4. yue/nano_v3_sft_profiled_comp_coding_50tests.jsonl +3 -0
  5. zh-hk/nano_v3_sft_profiled_comp_coding_50tests.jsonl +3 -0
.gitattributes CHANGED
@@ -59,3 +59,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
  nano_v3_sft_profiled_workbench.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
 
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
  nano_v3_sft_profiled_workbench.jsonl filter=lfs diff=lfs merge=lfs -text
62
+ en/nano_v3_sft_profiled_comp_coding_50tests.jsonl filter=lfs diff=lfs merge=lfs -text
63
+ yue/nano_v3_sft_profiled_comp_coding_50tests.jsonl filter=lfs diff=lfs merge=lfs -text
64
+ zh-hk/nano_v3_sft_profiled_comp_coding_50tests.jsonl filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -13,10 +13,11 @@ The original NVIDIA dataset blends multiple sources. For this parallel corpus, w
13
 
14
  | Subset / Source Dataset | Target Filename | English (`en`) | Cantonese (`yue`) | Traditional Chinese (`zh-hk`) |
15
  | :--- | :--- | :---: | :---: | :---: |
16
- | **BytedTsinghua-SIA/DAPO-Math-17k** (0.10 ratio) | `nano_v3_sft_profiled_dapo17k.jsonl` | 2,572 | 2,534 | 2,542 |
17
- | **Skywork/Skywork-OR1-RL-Data** (0.12 ratio) | `nano_v3_sft_profiled_skywork_no_omni.jsonl` | 2,998 | 2,975 | 2,979 |
18
- | **nvidia/Nemotron-RL-knowledge-mcqa** (0.20 ratio) | `nano_v3_sft_profiled_stem_mcqa.jsonl` | 2,639 | 2,850 | 2,690 |
19
- | **nvidia/Nemotron-RL-agent-workplace_assistant** (0.20 ratio) | `nano_v3_sft_profiled_workbench.jsonl` | 990 | 990 | 990 |
 
20
 
21
  ## Languages
22
  For each subset, files are provided in three parallel directories:
@@ -38,3 +39,12 @@ The original `nvidia/Nemotron-RL-agent-workplace_assistant` dataset features off
38
 
39
  To optimize this subset:
40
  - **Deduplication**: The raw dataset contained many duplicate user queries. We deduplicated the subset based on the user prompt, retaining only the **990 unique prompts** (and 1 system prompt) in the final parallel release.
 
 
 
 
 
 
 
 
 
 
13
 
14
  | Subset / Source Dataset | Target Filename | English (`en`) | Cantonese (`yue`) | Traditional Chinese (`zh-hk`) |
15
  | :--- | :--- | :---: | :---: | :---: |
16
+ | **[BytedTsinghua-SIA/DAPO-Math-17k](https://huggingface.co/datasets/BytedTsinghua-SIA/DAPO-Math-17k)** (0.10 ratio) | `nano_v3_sft_profiled_dapo17k.jsonl` | 2,572 | 2,534 | 2,542 |
17
+ | **[Skywork/Skywork-OR1-RL-Data](https://huggingface.co/datasets/Skywork/Skywork-OR1-RL-Data)** (0.12 ratio) | `nano_v3_sft_profiled_skywork_no_omni.jsonl` | 2,998 | 2,975 | 2,979 |
18
+ | **[nvidia/Nemotron-RL-knowledge-mcqa](https://huggingface.co/datasets/nvidia/Nemotron-RL-knowledge-mcqa)** (0.20 ratio) | `nano_v3_sft_profiled_stem_mcqa.jsonl` | 2,639 | 2,850 | 2,690 |
19
+ | **[nvidia/Nemotron-RL-agent-workplace_assistant](https://huggingface.co/datasets/nvidia/Nemotron-RL-agent-workplace_assistant)** (0.20 ratio) | `nano_v3_sft_profiled_workbench.jsonl` | 990 | 990 | 990 |
20
+ | **[nvidia/Nemotron-RL-coding-competitive_coding](https://huggingface.co/datasets/nvidia/Nemotron-RL-coding-competitive_coding)** (0.25 ratio) | `nano_v3_sft_profiled_comp_coding_50tests.jsonl` | 4,812 | 4,668 | 4,668 |
21
 
22
  ## Languages
23
  For each subset, files are provided in three parallel directories:
 
39
 
40
  To optimize this subset:
41
  - **Deduplication**: The raw dataset contained many duplicate user queries. We deduplicated the subset based on the user prompt, retaining only the **990 unique prompts** (and 1 system prompt) in the final parallel release.
42
+
43
+ ## nvidia/Nemotron-RL-coding-competitive_coding Processing
44
+ The `nvidia/Nemotron-RL-coding-competitive_coding` subset consists of competitive programming questions.
45
+
46
+ To process and translate this subset:
47
+ - **Deduplication**: We extracted the core coding questions and deduplicated them down to **4,291 unique questions** (out of 4,812 total records) to optimize translation.
48
+ - **Cantonese Translation**: The Cantonese translations were generated using the in-development model [jed351/cantonese_llm_v2_chat_gsm8k_GRPO_Qwen3-30B-A3B-Think-2507](https://huggingface.co/jed351/cantonese_llm_v2_chat_gsm8k_GRPO_Qwen3-30B-A3B-Think-2507).
49
+ - **Traditional Chinese Translation**: The Cantonese translations of the core questions were further translated into Traditional Chinese (`zh-hk`) using the [google/gemma-4-26B-A4B-it](https://huggingface.co/google/gemma-4-26B-A4B-it) model on vLLM.
50
+ - **Token Limit Filter**: We filtered out and excluded **128 unique questions** (mapping to 144 records in total) whose Cantonese translations exceeded the 16,384 token limit (usually caused by infinite repetition loops), resulting in exactly **4,668 parallel records** for both `yue` and `zh-hk` datasets.
en/nano_v3_sft_profiled_comp_coding_50tests.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:798d76b99ec9844424142636cab7567234b539f14e057621754fc57d47479314
3
+ size 1485145133
yue/nano_v3_sft_profiled_comp_coding_50tests.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96e7579e066acd71726e895f0d781766cd81b4b792941e834f5c21bf7658c9d5
3
+ size 1480652127
zh-hk/nano_v3_sft_profiled_comp_coding_50tests.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36ba8b3676f261cbc424d540f2cae140bc87cb27ff4be8bc98ec10275883ca51
3
+ size 1480745882