Add FP8 (dynamic) quantization for CodeLlama-7b-hf

README.md

@@ -0,0 +1,68 @@
+---
+license: other
+base_model: codellama/CodeLlama-7b-hf
+base_model_relation: quantized
+library_name: transformers
+pipeline_tag: text-generation
+tags:
+- fp8
+- compressed-tensors
+- vllm
+- quantized
+quantized_by: liodon-ai
+---
+
+# CodeLlama-7b-hf — FP8 (dynamic)
+
+FP8 quantization of [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf), published by [Liodon AI](https://huggingface.co/liodon-ai).
+
+Quantized with [llm-compressor](https://github.com/vllm-project/llm-compressor) using the
+`FP8_DYNAMIC` scheme: weights are cast to FP8 (E4M3) per-channel ahead of time, activations are
+quantized to FP8 dynamically per-token at inference time. No calibration dataset is needed for this
+scheme, so the quantized weights are numerically just a direct cast of the original — no calibration-set
+bias to worry about. `lm_head` is left unquantized (standard practice — negligible size, disproportionate
+quality impact if quantized).
+
+Original size: 13.5 GB → Quantized: 7.0 GB.
+
+## Quick Start
+
+**vLLM**
+```bash
+vllm serve liodon-ai/CodeLlama-7b-hf-FP8
+```
+
+**Text Generation Inference (TGI)**
+```bash
+docker run --gpus all -p 8080:80 ghcr.io/huggingface/text-generation-inference \
+    --model-id liodon-ai/CodeLlama-7b-hf-FP8
+```
+
+**SGLang**
+```bash
+python -m sglang.launch_server --model-path liodon-ai/CodeLlama-7b-hf-FP8
+```
+
+FP8 execution requires an NVIDIA GPU with compute capability ≥ 8.9 (Ada/Hopper/Blackwell — RTX 40-series,
+L4/L40S, H100/H200, B100/B200/GB10). On older GPUs, vLLM/TGI will dequantize to run, which loses the
+speed/memory benefit.
+
+## Source
+
+- **Model**: [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf)
+- **License**: other
+
+## Citation
+
+```bibtex
+@misc{liodonai_codellama_7b_hf_fp8,
+  title        = {CodeLlama-7b-hf — FP8},
+  author       = {{Liodon AI}},
+  year         = {2026},
+  howpublished = {\url{https://huggingface.co/liodon-ai/CodeLlama-7b-hf-FP8}},
+  note         = {FP8 (dynamic) quantization of codellama/CodeLlama-7b-hf}
+}
+```
+
+---
+*Quantized by [Liodon AI](https://huggingface.co/liodon-ai)*

config.json

@@ -0,0 +1,82 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 16384,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pad_token_id": null,
+  "pretraining_tp": 1,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "float-quantized",
+        "input_activations": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": true,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": null,
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "token",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        },
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "channel",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "float-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "lm_head"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.18.0"
+  },
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "rope_theta": 1000000,
+    "rope_type": "default"
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.14.1",
+  "use_cache": true,
+  "vocab_size": 32016
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "5.14.1"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f39ffdce61199e5759152f91a9b35f152890a66d9b9cf8fc62a01215ed853cfc
+size 7003868416

recipe.yaml

@@ -0,0 +1,8 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      targets: [Linear]
+      ignore: [lm_head]
+      scheme: FP8_DYNAMIC
+      bypass_divisibility_checks: false
+      requires_calibration_data: false

tokenizer_config.json

@@ -0,0 +1,28 @@
+{
+  "add_prefix_space": true,
+  "backend": "tokenizers",
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "eot_token": "▁<EOT>",
+  "extra_special_tokens": [
+    "▁<PRE>",
+    "▁<MID>",
+    "▁<SUF>",
+    "▁<EOT>",
+    "<FILL_ME>"
+  ],
+  "fill_token": "<FILL_ME>",
+  "is_local": false,
+  "legacy": null,
+  "local_files_only": false,
+  "middle_token": "▁<MID>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "prefix_token": "▁<PRE>",
+  "sp_model_kwargs": {},
+  "suffix_token": "▁<SUF>",
+  "tokenizer_class": "CodeLlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}