Initial Upload

README.md

@@ -0,0 +1,84 @@
+---
+license: other
+license_name: falcon-llm-license
+license_link: https://falconllm.tii.ae/falcon-terms-and-conditions.html
+language:
+  - en
+  - fr
+  - es
+  - pt
+pipeline_tag: text-generation
+tags:
+- causal-lm
+- autoround
+- auto-round
+- intel-autoround
+- woq
+- intel
+- pytorch
+- falcon3
+model_name: Falcon3 3B Instruct
+base_model:
+  - tiiuae/Falcon3-3B-Instruct
+inference: false
+library_name: transformers
+model_creator: tiiuae
+prompt_template: '{prompt} '
+quantized_by: fbaldassarri
+---
+
+## Model Information
+
+Quantized version of [tiiuae/Falcon3-3B-Instruct](https://huggingface.co/tiiuae/Falcon3-3B-Instruct) using torch.float32 for quantization tuning.
+- 4 bits (INT4)
+- group size = 128
+- Asymmetrical Quantization
+- Method WoQ (AutoRound format)
+
+Fast and low memory, 2-3X speedup (slight accuracy drop at W4G128)
+
+Quantization framework: [Intel AutoRound](https://github.com/intel/auto-round) v0.4.4
+
+Note: this INT4 version of Falcon3-3B-Instruct has been quantized to run inference through CPU.
+
+## Replication Recipe
+
+### Step 1 Install Requirements
+
+I suggest to install requirements into a dedicated python-virtualenv or a conda enviroment. 
+
+```
+wget https://github.com/intel/auto-round/archive/refs/tags/v0.4.4.tar.gz
+tar -xvzf v0.4.4.tar.gz
+cd auto-round-0.4.4
+pip install -r requirements-cpu.txt --upgrade
+```
+
+### Step 2 Build Intel AutoRound wheel from sources
+
+```
+pip install -vvv --no-build-isolation -e .[cpu]
+```
+
+### Step 3 Script for Quantization
+
+```
+  from transformers import AutoModelForCausalLM, AutoTokenizer
+  model_name = "tiiuae/Falcon3-3B-Instruct"
+  model = AutoModelForCausalLM.from_pretrained(model_name)
+  tokenizer = AutoTokenizer.from_pretrained(model_name)
+  from auto_round import AutoRound
+  bits, group_size, sym, device, amp = 4, 128, False, 'cpu', False
+  autoround = AutoRound(model, tokenizer, nsamples=128, iters=200, seqlen=512, batch_size=4, bits=bits, group_size=group_size, sym=sym, device=device, amp=amp)
+  autoround.quantize()
+  output_dir = "./AutoRound/tiiuae_Falcon3-3B-Instruct-autoround-int4-gs128-asym"
+  autoround.save_quantized(output_dir, format='auto_round', inplace=True)
+```
+
+## License
+
+[Falcon3 License](https://falconllm.tii.ae/falcon-terms-and-conditions.html)
+
+## Disclaimer
+
+This quantized model comes with no warranty. It has been developed only for research purposes. 

config.json

@@ -0,0 +1,54 @@
+{
+  "_name_or_path": "tiiuae/Falcon3-3B-Instruct",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 11,
+  "head_dim": 256,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 9216,
+  "max_position_embeddings": 32768,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 22,
+  "num_key_value_heads": 4,
+  "pretraining_tp": 1,
+  "quantization_config": {
+    "amp": false,
+    "autoround_version": "0.4.4",
+    "backend": "auto_round:exllamav2",
+    "batch_size": 4,
+    "bits": 4,
+    "data_type": "int",
+    "dataset": "NeelNanda/pile-10k",
+    "enable_minmax_tuning": true,
+    "enable_norm_bias_tuning": false,
+    "enable_quanted_input": true,
+    "gradient_accumulate_steps": 1,
+    "group_size": 128,
+    "iters": 200,
+    "low_gpu_mem_usage": false,
+    "lr": 0.005,
+    "minmax_lr": 0.005,
+    "nsamples": 128,
+    "quant_method": "intel/auto-round",
+    "scale_dtype": "torch.float16",
+    "seqlen": 512,
+    "sym": false,
+    "to_quant_block_names": null
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000042,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.0",
+  "use_cache": true,
+  "vocab_size": 131072
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 11,
+  "eos_token_id": 11,
+  "transformers_version": "4.48.0"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b5d985f721779ec4e3ec0331d6452e2afd1e501a57eccf635b5fc1ad8b7d468
+size 4482700120

quantization_config.json

@@ -0,0 +1,24 @@
+{
+  "bits": 4,
+  "group_size": 128,
+  "sym": false,
+  "data_type": "int",
+  "enable_quanted_input": true,
+  "enable_minmax_tuning": true,
+  "seqlen": 512,
+  "batch_size": 4,
+  "scale_dtype": "torch.float16",
+  "lr": 0.005,
+  "minmax_lr": 0.005,
+  "gradient_accumulate_steps": 1,
+  "iters": 200,
+  "amp": false,
+  "nsamples": 128,
+  "low_gpu_mem_usage": false,
+  "to_quant_block_names": null,
+  "enable_norm_bias_tuning": false,
+  "dataset": "NeelNanda/pile-10k",
+  "autoround_version": "0.4.4",
+  "quant_method": "intel/auto-round",
+  "backend": "auto_round:exllamav2"
+}

special_tokens_map.json

@@ -0,0 +1,41 @@
+{
+  "additional_special_tokens": [
+    ">>TITLE<<",
+    ">>ABSTRACT<<",
+    ">>INTRODUCTION<<",
+    ">>SUMMARY<<",
+    ">>COMMENT<<",
+    ">>ANSWER<<",
+    ">>QUESTION<<",
+    ">>DOMAIN<<",
+    ">>EMAIL_ADDRESS<<",
+    ">>IP_ADDRESS<<",
+    "<|startoftext|>",
+    ">>IP_ADDRESS_0<<",
+    ">>IP_ADDRESS_1<<",
+    ">>IP_ADDRESS_2<<",
+    ">>IP_ADDRESS_3<<",
+    ">>IP_ADDRESS_4<<",
+    ">>IP_ADDRESS_5<<",
+    ">>IP_ADDRESS_6<<",
+    ">>IP_ADDRESS_7<<",
+    ">>IP_ADDRESS_8<<",
+    ">>IP_ADDRESS_9<<",
+    ">>PASSWORD<<",
+    ">>KEY<<"
+  ],
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}