init: release

README.md

@@ -1,132 +1,55 @@
 ---
-license: cc-by-sa-4.0
-datasets:
-- tiiuae/falcon-refinedweb
-- togethercomputer/RedPajama-Data-1T
-- uonlp/CulturaX
-- CarperAI/pilev2-dev
-- bigcode/starcoderdata
-- DataProvenanceInitiative/Commercially-Verified-Licenses
+license: other
 language:
 - en
 tags:
 - causal-lm
 ---
-# `StableLM-1.6B`
+# `Stable LM 2 1.6B` (global_step420000)
 
-## Model Description
+## Description
 
-`StableLM-1.6B` is a 1.6 billion parameter decoder-only language model pre-trained on 2 trillion tokens of diverse multi-lingual and code datasets for two epochs.
+`Stable LM 2 1.6B` is a 1.6 billion parameter decoder-only language model pre-trained on 2 trillion tokens of diverse multilingual and code datasets for two epochs.
 
 ## Usage
 
-Get started generating text with `StableLM-1.6B` by using the following code snippet:
+This branch contains the training checkpoint for `Stable LM 2 1.6B` at step 420,000. It is the final checkpoint taken before cooldown.
+We provide the following contents in the [`global_step420000`](https://huggingface.co/stabilityai/stablelm-2-1_6b/tree/global_step420000/global_step420000) directory:
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-1_6b", trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained(
-  "stabilityai/stablelm-1_6b",
-  trust_remote_code=True,
-  torch_dtype="auto",
-)
-model.cuda()
-inputs = tokenizer("The weather is always wonderful", return_tensors="pt").to(model.device)
-tokens = model.generate(
-  **inputs,
-  max_new_tokens=64,
-  temperature=0.70,
-  top_p=0.95,
-  do_sample=True,
-)
-print(tokenizer.decode(tokens[0], skip_special_tokens=True))
-```
+- `bf16_zero_pp_mp_rank_00_optim_states.pt`: The Adam states and FP32 weights for each parameter. You will need to port this to your optimizer format when importing into your training process.
+
+- `mp_rank_00_model_states.pt`: The model weights following the [GPT-NeoX](https://github.com/EleutherAI/gpt-neox) convention.
 
-### Run with Flash Attention 2 ⚡️
+- `config.yml`: The pre-training configuration file for this checkpoint. Linear learning rate cooldown should be taken from `lr=0.0002529` to `lr=0.0`.
 
-<details>
-<summary> Click to expand </summary>
+The model weights are also converted to HuggingFace `transformers` format and can be loaded with the following code:
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
-tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-1_6b", trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-2-1_6b", trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
-  "stabilityai/stablelm-1_6b",
+  "stabilityai/stablelm-2-1_6b",
   trust_remote_code=True,
   torch_dtype="auto",
-  attn_implementation="flash_attention_2",
+  revision="global_step420000"
 )
 model.cuda()
-inputs = tokenizer("The weather is always wonderful", return_tensors="pt").to(model.device)
-tokens = model.generate(
-  **inputs,
-  max_new_tokens=64,
-  temperature=0.70,
-  top_p=0.95,
-  do_sample=True,
-)
-print(tokenizer.decode(tokens[0], skip_special_tokens=True))
 ```
 
-</details>
-
-
-## Model Details
-
-* **Developed by**: [Stability AI](https://stability.ai/)
-* **Model type**: `StableLM-1.6B` models are auto-regressive language models based on the transformer decoder architecture.
-* **Language(s)**: English
-* **Library**: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
-* **License**: Model checkpoints are licensed under the Creative Commons license ([CC BY-SA-4.0](https://creativecommons.org/licenses/by-sa/4.0/)). Under this license, you must give [credit](https://creativecommons.org/licenses/by/4.0/#) to Stability AI, provide a link to the license, and [indicate if changes were made](https://creativecommons.org/licenses/by/4.0/#). You may do so in any reasonable manner, but not in any way that suggests the Stability AI endorses you or your use.
-* **Contact**: For questions and comments about the model, please email `lm@stability.ai`
-
-### Model Architecture
+## License
 
-The model is a decoder-only transformer similar to the LLaMA ([Touvron et al., 2023](https://arxiv.org/abs/2307.09288)) architecture with the following modifications:
+* **Stability AI Non-Commercial Research Community License**. If you'd like to use this model for commercial products or purposes, please contact us [here](https://stability.ai/membership) to learn more.
 
-| Parameters     | Hidden Size | Layers | Heads | Sequence Length |
-|----------------|-------------|--------|-------|-----------------|
-| 1,644,417,024  | 2048        | 24     | 32    | 4096            |
+## Acknowledgements
 
-* **Position Embeddings**: Rotary Position Embeddings ([Su et al., 2021](https://arxiv.org/abs/2104.09864)) applied to the first 25% of head embedding dimensions for improved throughput following [Black et al. (2022)](https://arxiv.org/pdf/2204.06745.pdf).
-* **Normalization**: LayerNorm ([Ba et al., 2016](https://arxiv.org/abs/1607.06450)) with learned bias terms as opposed to RMSNorm ([Zhang & Sennrich, 2019](https://arxiv.org/abs/1910.07467)).
-* **Biases**: We remove all bias terms from the model except for attention Q,K,V projections ([Bai et al., 2023](https://arxiv.org/abs/2309.16609)).
-* **Tokenizer**: We use Arcade100k, a BPE tokenizer extended from OpenAI's [`tiktoken.cl100k_base`](https://github.com/openai/tiktoken). We split digits into individual tokens following findings by [Liu & Low (2023)](https://arxiv.org/abs/2305.14201).
+- Dakota Mahan for creating the ZeRO optimizer state merging script.
 
-## Training
-
-### Training Dataset
-
-The dataset is comprised of a filtered mixture of open-source large-scale datasets available on the [HuggingFace Hub](https://huggingface.co/datasets): Falcon RefinedWeb extract ([Penedo et al., 2023](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)), RedPajama-Data ([Together Computer., 2023](https://github.com/togethercomputer/RedPajama-Data)) and The Pile ([Gao et al., 2020](https://arxiv.org/abs/2101.00027)) both without the *Books3* subset, and StarCoder ([Li et al., 2023](https://arxiv.org/abs/2305.06161)). We further supplement our training with multi-lingual data from CulturaX ([Nguyen et al., 2023](https://arxiv.org/abs/2309.09400)) and, in particular, from its OSCAR corpora, as well as restructured data in the style of [Yuan & Liu (2022)](https://arxiv.org/abs/2206.11147).
-
-* Given the large amount of web data, we recommend fine-tuning the base StableLM-1.6B for your downstream tasks.
-
-### Training Procedure
-
-The model is pre-trained on the aforementioned datasets in `bfloat16` precision, optimized with AdamW, and trained using the NeoX tokenizer with a vocabulary size of 100,352. We outline the complete hyperparameters choices in the project's [GitHub repository - config (TODO)](https://github.com/Stability-AI/StableLM/blob/main/configs/stablelm-1.6b.yml).
-
-### Training Infrastructure
-
-* **Hardware**: `StableLM-1.6B` was trained on the Stability AI cluster across 512 NVIDIA A100 40GB GPUs (AWS P4d instances).
-
-* **Software**: We use a fork of `gpt-neox` ([EleutherAI, 2021](https://github.com/EleutherAI/gpt-neox)), train under 2D parallelism (Data and Tensor Parallel) with ZeRO-1 ([Rajbhandari et al., 2019](https://arxiv.org/abs/1910.02054v3)), and rely on flash-attention as well as SwiGLU and Rotary Embedding kernels from FlashAttention-2 ([Dao et al., 2023](https://tridao.me/publications/flash2/flash2.pdf))
-
-## Use and Limitations
-
-### Intended Use
-
-The model is intended to be used as a foundational base model for application-specific fine-tuning. Developers must evaluate and fine-tune the model for safe performance in downstream applications.
-
-### Limitations and Bias
-​
-As a base model, this model may exhibit unreliable, unsafe, or other undesirable behaviors that must be corrected through evaluation and fine-tuning prior to deployment. The pre-training dataset may have contained offensive or inappropriate content, even after applying data cleansing filters, which can be reflected in the model-generated text. We recommend that users exercise caution when using these models in production systems. Do not use the models if they are unsuitable for your application, or for any applications that may cause deliberate or unintentional harm to others.
-
-## How to Cite
+## Citation
 
 ```bibtex
-@misc{StableLM-1.6B,
-      url={[https://huggingface.co/stabilityai/stablelm-1.6b](https://huggingface.co/stabilityai/stablelm-1.6b)},
-      title={StableLM 1.6B},
+@misc{StableLM-2-1.6B,
+      url={[https://huggingface.co/stabilityai/stablelm-2-1_6b](https://huggingface.co/stabilityai/stablelm-2-1_6b)},
+      title={Stable LM 2 1.6B},
       author={Stability AI Language Team}
 }
-```
+```

generation_config.json

@@ -2,5 +2,5 @@
   "_from_model_config": true,
   "bos_token_id": 100257,
   "eos_token_id": 100257,
-  "transformers_version": "4.36.2"
+  "transformers_version": "4.35.2"
 }

global_step420000/bf16_zero_pp_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f24d0cd5a8289d48d9f59899ca029a17ced8257c81403e5fb0bee95ded3bb81
+size 19733203617

global_step420000/config.yml

@@ -0,0 +1,139 @@
+{
+  # parallelism settings
+  "pipe-parallel-size": 0,
+  "model-parallel-size": 1,
+
+  # model settings
+  "num-layers": 24,
+  "hidden-size": 2048,
+  "num-attention-heads": 32,
+  "seq-length": 4096,
+  "max-position-embeddings": 4096,
+
+  # architecture design
+  "attention_head_type": "multihead",
+  "norm": "layernorm",
+  "pos-emb": "rotary",
+  "rotary_pct": 0.25,
+  "rotary_interleaved": false,  # GPT-NeoX style
+  "mlp_multiple_of": 256,
+  "mlp_type": "gated",
+  "activation": "silu",
+  "no-weight-tying": true,
+  "gpt_j_residual": false,
+  "gpt_j_tied": false,
+  "output_layer_parallelism": "column",
+
+  # init methods
+  "init_method": "normal",
+  "output_layer_init_method": "scaled_normal",
+  "init_method_std": 0.02,
+
+  # biases
+  # NOTE: QKV projections were hard-coded with bias=True
+  "use_bias_in_norms": false,
+  "use_bias_in_attn_linear": false,
+  "use_bias_in_mlp": false,
+
+  # fused ops
+  "use_flash_cross_entropy": true,
+  "bias-gelu-fusion": false,
+  "scaled-upper-triang-masked-softmax-fusion": false,
+  "attention-config": [[["flash"], 24]],
+
+  # optimizer settings
+  "optimizer": {
+    "type": "Adam",
+    "params": {
+      "lr": 0.001,
+      "betas": [0.9, 0.95],
+      "eps": 1.0e-8,
+    }
+  },
+  "min_lr": 0.0001,
+  "train-iters": 540_000,
+  "lr-decay-iters": 540_000,
+  "lr-decay-style": "hybrid_cosine_inv_sqrt_2",
+  "warmup": 0.018,
+  "cooldown": 0.,
+
+  "reset_attention_mask": true,
+  "reset_position_ids": true,
+
+  # for all zero_optimization options, see https://www.deepspeed.ai/docs/config-json/#zero-optimizations-for-fp16-training
+  "zero_optimization": {
+    "stage": 1,
+    "allgather_partitions": true,
+    "allgather_bucket_size": 1260000000,
+    "overlap_comm": true,
+    "reduce_scatter": true,
+    "reduce_bucket_size": 1260000000,
+    "contiguous_gradients": true,
+    "cpu_offload": false,
+  },
+
+  # batch / data settings
+  "train_micro_batch_size_per_gpu": 2,
+  "gradient_accumulation_steps": 2,
+  "data-impl": "mmap",
+  "eval-interval": 500_000,
+  "eval-iters": 1,
+  "eval_batch_size": 1,
+  "eval_tasks": [],
+
+  # activation checkpointing
+  "checkpoint-activations": true,
+  "checkpoint-num-layers": 24,
+  "partition-activations": true,
+  "synchronize-each-layer": true,
+
+  # regularization
+  "gradient_clipping": 1,
+  "weight-decay": 0.1,
+  "hidden-dropout": 0.,
+  "attention-dropout": 0.,
+
+  # precision settings
+  "bf16": { "enabled": true },
+  "precision": "bfloat16",
+  "full_precision_lm_cross_entropy": true,
+  "fp32_allreduce": true,
+
+  # misc. training settings
+  "num-workers": 2,
+  "distributed-backend": "nccl",
+
+  # checkpoint settings
+  "checkpoint-factor": 10_000,
+  #"s3_sync_interval": 20_000,
+  "extra-save-iters": [230_001],
+  "save": "",
+  "load": "",
+  #"s3_path": "",
+
+  "train_data_paths": [],
+  "valid-data-paths": [],
+  "test-data-paths": [],
+
+  # tokenizer settings
+  "tokenizer-type": "TiktokenTokenizer",
+  "vocab-file": "arcade100k.tiktoken",
+
+  "log-interval": 10,
+  "steps_per_print": 10,
+  "wall_clock_breakdown": true,
+
+  "use_wandb": true,
+  "wandb_host": "",
+  "wandb_team": "",
+  "wandb_project": "",
+  "wandb_group": "",
+  "wandb_name": "",
+  #"wandb_resume": "must",
+
+  # multi-node launcher
+  "launcher": "slurm",
+  "deepspeed_slurm": true,
+
+  "seed": 1234
+}

global_step420000/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04a05847e6c4a5968e8780b1a0895051c6e1666e3912842ece4f70cddf6395b9
+size 3288906844

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8bdf317e2b35ab5c8009cbb6c7ce495e4e608a6b9b843d44054edf25b8c5860d
+oid sha256:f4cf5c39e33b678995a8b593c1670feed04213371e1840d07b3cdde950a309c7
 size 3289069520