Replace model with updated FP8 dynamic quantization

Removing all existing files and uploading new FP8 dynamic quantized model weights

README.md

@@ -0,0 +1,334 @@
+---
+license: apache-2.0
+pipeline_tag: text-generation
+tags:
+- fp8
+- quantized
+- llm-compressor
+- compressed-tensors
+- red hat
+base_model:
+- ibm-granite/granite-4.0-h-small
+---
+
+
+# Granite-4.0-h-small
+
+## Model Overview
+- **Model Architecture:** GraniteMoeHybridForCausalLM
+  - **Input:** Text
+  - **Output:** Text
+- **Model Optimizations:**
+  - **Weight quantization:** FP8
+  - **Activation quantization:** FP8
+- **Release Date:** 
+- **Version:** 1.0
+- **Model Developers:**: Red Hat
+
+Quantized version of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights and activations of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small) to FP8 data type.
+This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
+Only the weights and activations of the linear operators within transformers blocks of the language model are quantized. 
+
+## Deployment
+
+### Use with vLLM
+
+1. Install vLLM from main:
+```
+uv pip install -U git+https://github.com/vllm-project/vllm.git \
+  --extra-index-url https://wheels.vllm.ai/nightly \
+  --no-deps \
+  --no-cache
+
+
+uv pip install compressed-tensors==0.12.3a20251114 --no-cache
+uv pip install --upgrade torchvision --break-system-packages --no-cache
+uv pip install cloudpickle msgspec zmq blake3 cachetools prometheus_client fastapi openai openai_harmony pybase64 llguidance diskcache xgrammar lm-format-enforcer partial-json-parser cbor2 einops gguf numba --no-cache
+```
+
+2. Initialize vLLM server:
+```
+vllm serve RedHatAI/granite-4.0-h-small-FP8-dynamic --tensor_parallel_size 1
+```
+
+3. Send requests to the server:
+
+```python
+from openai import OpenAI
+
+# Modify OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://<your-server-host>:8000/v1"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+model = "RedHatAI/granite-4.0-h-small-FP8-dynamic"
+
+messages = [
+    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
+]
+
+
+outputs = client.chat.completions.create(
+    model=model,
+    messages=messages,
+)
+
+generated_text = outputs.choices[0].message.content
+print(generated_text)
+```
+
+## Creation
+
+This model was quantized using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as shown below.
+
+
+<details>
+  <summary>Creation details</summary>
+
+Install specific llm-compression version:
+```
+uv pip install git+https://github.com/vllm-project/llm-compressor.git@refs/pull/2001/head --no-cache
+uv pip install --upgrade torchvision --break-system-packages --no-cache
+```
+
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from llmcompressor import oneshot
+from llmcompressor.modifiers.quantization import QuantizationModifier
+from llmcompressor.utils import dispatch_for_generation
+from llmcompressor.modeling import replace_modules_for_calibration
+from llmcompressor.modeling.granite4 import pack_3d_experts
+
+
+MODEL_ID = "ibm-granite/granite-4.0-h-small"
+
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+
+model = replace_modules_for_calibration(model)
+
+ignore_lay = ["lm_head", "re:.*block_sparse_moe.router"]
+
+recipe = QuantizationModifier(
+    targets=["Linear"],
+    scheme="FP8_DYNAMIC",
+    ignore=ignore_lay,
+)
+
+oneshot(model=model, recipe=recipe)
+
+print("========== SAMPLE GENERATION ==============")
+dispatch_for_generation(model)
+input_ids = tokenizer(
+    "Describe Large Language Model", return_tensors="pt"
+).input_ids.to(model.device)
+output = model.generate(input_ids, max_new_tokens=35)
+print(tokenizer.decode(output[0]))
+print("==========================================")
+
+SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-dynamic"
+print(f"Saving to {SAVE_DIR}")
+
+model.save_pretrained(SAVE_DIR)
+tokenizer.save_pretrained(SAVE_DIR)
+pack_3d_experts(SAVE_DIR)
+```
+</details>
+
+
+## Evaluation
+
+
+The model was evaluated on the OpenLLM leaderboard task, using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness).
+[vLLM](https://docs.vllm.ai/en/stable/) was used for all evaluations.
+
+<details>
+  <summary>Evaluation details</summary>
+
+  Install vLLM from main:
+```
+uv pip install -U git+https://github.com/vllm-project/vllm.git \
+  --extra-index-url https://wheels.vllm.ai/nightly \
+  --no-deps \
+  --no-cache
+
+
+uv pip install compressed-tensors==0.12.3a20251114 --no-cache
+uv pip install --upgrade torchvision --break-system-packages --no-cache
+uv pip install cloudpickle msgspec zmq blake3 cachetools prometheus_client fastapi openai openai_harmony pybase64 llguidance diskcache xgrammar lm-format-enforcer partial-json-parser cbor2 einops gguf numba --no-cache
+```
+  
+  **Openllm V1**
+  ```
+  lm_eval \
+    --model vllm \
+    --model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=16384,tensor_parallel_size=1,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
+    --tasks openllm \
+    --write_out \
+    --batch_size auto \
+    --show_config
+  ```
+
+
+  **Openllm V2**  
+  ```
+  lm_eval \
+    --model vllm \
+    --model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-dynamic",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=1,gpu_memory_utilization=0.7,disable_log_stats=True,enable_chunked_prefill=True,trust_remote_code=True \
+    --tasks leaderboard \
+    --apply_chat_template \
+    --fewshot_as_multiturn \
+    --write_out \
+    --batch_size auto \
+    --show_config
+  ```
+
+
+  **Coding Benchmarks**
+
+  ```
+  evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-dynamic" \
+                    --dataset "humaneval" \
+                    --backend vllm \
+                    --tp 1 \
+                    --greedy
+
+  evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-dynamic" \
+                  --dataset "mbpp" \
+                  --backend vllm \
+                  --tp 1 \
+                  --greedy
+
+  ```
+
+</details>
+
+
+
+### Accuracy Comparison
+<table>
+  <thead>
+    <tr>
+      <th>Category</th>
+      <th>Metric</th>
+      <th>ibm-granite/granite-4.0-h-small</th>
+      <th>ibm-granite/granite-4.0-h-small-FP8</th>
+      <th>RedHatAI/granite-4.0-h-small-FP8-block</th>
+      <th>RedHatAI/granite-4.0-h-small-FP8-dynamic</th>
+    </tr>
+  </thead>
+  <tbody>
+    <!-- OpenLLM Leaderboard V1 -->
+    <tr>
+      <td rowspan="7"><b>OpenLLM V1</b></td>
+      <td>ARC-Challenge (Acc-Norm, 25-shot)</td>
+      <td>72.27</td>
+      <td>72.10 (99.76%)</td>
+      <td>72.27 (100.00%)</td>
+      <td>72.10 (99.76%)</td>
+    </tr>
+    <tr>
+      <td>GSM8K (Strict-Match, 5-shot)</td>
+      <td>85.22</td>
+      <td>85.29 (100.09%)</td>
+      <td>85.52 (100.36%)</td>
+      <td>84.84 (99.56%)</td>
+    </tr>
+    <tr>
+      <td>HellaSwag (Acc-Norm, 10-shot)</td>
+      <td>86.08</td>
+      <td>85.88 (99.77%)</td>
+      <td>85.96 (99.86%)</td>
+      <td>85.88 (99.77%)</td>
+    </tr>
+    <tr>
+      <td>MMLU (Acc, 5-shot)</td>
+      <td>77.15</td>
+      <td>77.18 (100.03%)</td>
+      <td>77.23 (100.09%)</td>
+      <td>77.18 (100.03%)</td>
+    </tr>
+    <tr>
+      <td>TruthfulQA (MC2, 0-shot)</td>
+      <td>57.64</td>
+      <td>57.63 (99.99%)</td>
+      <td>57.94 (100.52%)</td>
+      <td>57.63 (100.00%)</td>
+    </tr>
+    <tr>
+      <td>Winogrande (Acc, 5-shot)</td>
+      <td>81.37</td>
+      <td>81.45 (100.10%)</td>
+      <td>80.82 (99.32%)</td>
+      <td>81.45 (100.10%)</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>76.62</b></td>
+      <td><b>76.59 (99.96%)</b></td>
+      <td><b>76.62 (100.00%)</b></td>
+      <td><b>76.51 (99.86%)</b></td>
+    </tr>
+    <!-- OpenLLM Leaderboard V2 -->
+    <tr>
+      <td rowspan="7"><b>OpenLLM V2</b></td>
+      <td>IFEval (Inst Level Strict Acc, 0-shot)</td>
+      <td>87.53</td>
+      <td>87.17 (99.59%)</td>
+      <td>86.69 (99.04%)</td>
+      <td>87.41 (99.86%)</td>
+    </tr>
+    <tr>
+      <td>BBH (Acc-Norm, 3-shot)</td>
+      <td>61.52</td>
+      <td>61.31 (99.66%)</td>
+      <td>61.40 (99.80%)</td>
+      <td>61.19 (99.46%)</td>
+    </tr>
+    <tr>
+      <td>Math-Hard (Exact-Match, 4-shot)</td>
+      <td>46.22</td>
+      <td>43.73 (94.61%)</td>
+      <td>43.88 (94.93%)</td>
+      <td>41.77 (90.36%)</td>
+    </tr>
+    <tr>
+      <td>GPQA (Acc-Norm, 0-shot)</td>
+      <td>35.23</td>
+      <td>34.98 (99.29%)</td>
+      <td>34.23 (97.14%)</td>
+      <td>34.23 (97.14%)</td>
+    </tr>
+    <tr>
+      <td>MUSR (Acc-Norm, 0-shot)</td>
+      <td>46.69</td>
+      <td>46.56 (99.72%)</td>
+      <td>45.77 (98.02%)</td>
+      <td>45.77 (98.02%)</td>
+    </tr>
+    <tr>
+      <td>MMLU-Pro (Acc, 5-shot)</td>
+      <td>47.99</td>
+      <td>47.63 (99.26%)</td>
+      <td>47.93 (99.88%)</td>
+      <td>47.58 (99.15%)</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>54.20</b></td>
+      <td><b>53.56 (98.82%)</b></td>
+      <td><b>53.32 (98.38%)</b></td>
+      <td><b>52.99 (97.77%)</b></td>
+    </tr>
+  </tbody>
+</table>

config.json

@@ -88,9 +88,11 @@
           "num_bits": 8,
           "observer": null,
           "observer_kwargs": {},
+          "scale_dtype": null,
           "strategy": "token",
           "symmetric": true,
-          "type": "float"
+          "type": "float",
+          "zp_dtype": null
         },
         "output_activations": null,
         "targets": [
@@ -102,11 +104,13 @@
           "dynamic": false,
           "group_size": null,
           "num_bits": 8,
-          "observer": "mse",
+          "observer": "minmax",
           "observer_kwargs": {},
+          "scale_dtype": null,
           "strategy": "channel",
           "symmetric": true,
-          "type": "float"
+          "type": "float",
+          "zp_dtype": null
         }
       }
     },
@@ -114,117 +118,45 @@
     "global_compression_ratio": null,
     "ignore": [
       "model.layers.0.block_sparse_moe.router.layer",
-      "model.layers.0.mamba.in_proj",
-      "model.layers.0.mamba.out_proj",
       "model.layers.1.block_sparse_moe.router.layer",
-      "model.layers.1.mamba.in_proj",
-      "model.layers.1.mamba.out_proj",
       "model.layers.2.block_sparse_moe.router.layer",
-      "model.layers.2.mamba.in_proj",
-      "model.layers.2.mamba.out_proj",
       "model.layers.3.block_sparse_moe.router.layer",
-      "model.layers.3.mamba.in_proj",
-      "model.layers.3.mamba.out_proj",
       "model.layers.4.block_sparse_moe.router.layer",
-      "model.layers.4.mamba.in_proj",
-      "model.layers.4.mamba.out_proj",
       "model.layers.5.block_sparse_moe.router.layer",
       "model.layers.6.block_sparse_moe.router.layer",
-      "model.layers.6.mamba.in_proj",
-      "model.layers.6.mamba.out_proj",
       "model.layers.7.block_sparse_moe.router.layer",
-      "model.layers.7.mamba.in_proj",
-      "model.layers.7.mamba.out_proj",
       "model.layers.8.block_sparse_moe.router.layer",
-      "model.layers.8.mamba.in_proj",
-      "model.layers.8.mamba.out_proj",
       "model.layers.9.block_sparse_moe.router.layer",
-      "model.layers.9.mamba.in_proj",
-      "model.layers.9.mamba.out_proj",
       "model.layers.10.block_sparse_moe.router.layer",
-      "model.layers.10.mamba.in_proj",
-      "model.layers.10.mamba.out_proj",
       "model.layers.11.block_sparse_moe.router.layer",
-      "model.layers.11.mamba.in_proj",
-      "model.layers.11.mamba.out_proj",
       "model.layers.12.block_sparse_moe.router.layer",
-      "model.layers.12.mamba.in_proj",
-      "model.layers.12.mamba.out_proj",
       "model.layers.13.block_sparse_moe.router.layer",
-      "model.layers.13.mamba.in_proj",
-      "model.layers.13.mamba.out_proj",
       "model.layers.14.block_sparse_moe.router.layer",
-      "model.layers.14.mamba.in_proj",
-      "model.layers.14.mamba.out_proj",
       "model.layers.15.block_sparse_moe.router.layer",
       "model.layers.16.block_sparse_moe.router.layer",
-      "model.layers.16.mamba.in_proj",
-      "model.layers.16.mamba.out_proj",
       "model.layers.17.block_sparse_moe.router.layer",
-      "model.layers.17.mamba.in_proj",
-      "model.layers.17.mamba.out_proj",
       "model.layers.18.block_sparse_moe.router.layer",
-      "model.layers.18.mamba.in_proj",
-      "model.layers.18.mamba.out_proj",
       "model.layers.19.block_sparse_moe.router.layer",
-      "model.layers.19.mamba.in_proj",
-      "model.layers.19.mamba.out_proj",
       "model.layers.20.block_sparse_moe.router.layer",
-      "model.layers.20.mamba.in_proj",
-      "model.layers.20.mamba.out_proj",
       "model.layers.21.block_sparse_moe.router.layer",
-      "model.layers.21.mamba.in_proj",
-      "model.layers.21.mamba.out_proj",
       "model.layers.22.block_sparse_moe.router.layer",
-      "model.layers.22.mamba.in_proj",
-      "model.layers.22.mamba.out_proj",
       "model.layers.23.block_sparse_moe.router.layer",
-      "model.layers.23.mamba.in_proj",
-      "model.layers.23.mamba.out_proj",
       "model.layers.24.block_sparse_moe.router.layer",
-      "model.layers.24.mamba.in_proj",
-      "model.layers.24.mamba.out_proj",
       "model.layers.25.block_sparse_moe.router.layer",
       "model.layers.26.block_sparse_moe.router.layer",
-      "model.layers.26.mamba.in_proj",
-      "model.layers.26.mamba.out_proj",
       "model.layers.27.block_sparse_moe.router.layer",
-      "model.layers.27.mamba.in_proj",
-      "model.layers.27.mamba.out_proj",
       "model.layers.28.block_sparse_moe.router.layer",
-      "model.layers.28.mamba.in_proj",
-      "model.layers.28.mamba.out_proj",
       "model.layers.29.block_sparse_moe.router.layer",
-      "model.layers.29.mamba.in_proj",
-      "model.layers.29.mamba.out_proj",
       "model.layers.30.block_sparse_moe.router.layer",
-      "model.layers.30.mamba.in_proj",
-      "model.layers.30.mamba.out_proj",
       "model.layers.31.block_sparse_moe.router.layer",
-      "model.layers.31.mamba.in_proj",
-      "model.layers.31.mamba.out_proj",
       "model.layers.32.block_sparse_moe.router.layer",
-      "model.layers.32.mamba.in_proj",
-      "model.layers.32.mamba.out_proj",
       "model.layers.33.block_sparse_moe.router.layer",
-      "model.layers.33.mamba.in_proj",
-      "model.layers.33.mamba.out_proj",
       "model.layers.34.block_sparse_moe.router.layer",
-      "model.layers.34.mamba.in_proj",
-      "model.layers.34.mamba.out_proj",
       "model.layers.35.block_sparse_moe.router.layer",
       "model.layers.36.block_sparse_moe.router.layer",
-      "model.layers.36.mamba.in_proj",
-      "model.layers.36.mamba.out_proj",
       "model.layers.37.block_sparse_moe.router.layer",
-      "model.layers.37.mamba.in_proj",
-      "model.layers.37.mamba.out_proj",
       "model.layers.38.block_sparse_moe.router.layer",
-      "model.layers.38.mamba.in_proj",
-      "model.layers.38.mamba.out_proj",
       "model.layers.39.block_sparse_moe.router.layer",
-      "model.layers.39.mamba.in_proj",
-      "model.layers.39.mamba.out_proj",
       "lm_head"
     ],
     "kv_cache_scheme": null,
@@ -232,7 +164,7 @@
     "quantization_status": "compressed",
     "sparsity_config": {},
     "transform_config": {},
-    "version": "0.12.3.a20251023"
+    "version": "0.12.3.a20251114"
   },
   "residual_multiplier": 0.22,
   "rms_norm_eps": 1e-05,
@@ -240,8 +172,8 @@
   "rope_theta": 10000,
   "router_aux_loss_coef": 0.0,
   "shared_intermediate_size": 1536,
-  "tie_word_embeddings": false,
-  "transformers_version": "4.56.2",
+  "tie_word_embeddings": true,
+  "transformers_version": "4.57.1",
   "use_cache": true,
   "vocab_size": 100352
 }

generation_config.json

@@ -3,5 +3,5 @@
   "bos_token_id": 100257,
   "eos_token_id": 100257,
   "pad_token_id": 100256,
-  "transformers_version": "4.56.2"
+  "transformers_version": "4.57.1"
 }

model-00001-of-00014.safetensors → model-00001-of-00007.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4a6e0834590ac7935a737d79be3aecdb85d7357d52b2310178bead818aba9e9
-size 4894104208
+oid sha256:617302bc883f67a43399c70fad10eb5c01e23e40eac7d69fead9a1617df0335c
+size 4664217352

model-00002-of-00014.safetensors → model-00002-of-00007.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2091101cbf3f58e83c1ae6f1aa1f9e03e50a3da3d66fa8ae7f920992e1ee6e1a
-size 4749076040
+oid sha256:c245088455a60f9dabc0fcc2d78d3cc8455e97fef3f5467a4d032891465e1fb5
+size 4664193144

model-00003-of-00014.safetensors → model-00003-of-00007.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e4e5a282df3e2e0500e270ef132f9cd72a26d51c569db047c172911b299b6c4
-size 4586465728
+oid sha256:8d7b0fdb19ba78a4a1a27722612ed91b168e5ad334f0433557fca759491e565b
+size 4654421800

model-00004-of-00014.safetensors → model-00004-of-00007.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:596fed8185c2cbd24dd639dee43c866b96a5667ba4c49ec97ace499a1aba4217
-size 4749076080
+oid sha256:78cd2af57c0f7a8fb6c1fa70a25746eef17a2078631a91b3113b2b4483093785
+size 4657942800

model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17701a6d57dc4a5c78dda6a26ea7939f2901042375b15ed966ce5afaaac3d435
+size 4648224424

model-00005-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4ef8a4778927e73f4d4b8ec1a60fddb9f236ed3074d8893fa423653274ac9205
-size 4749076088

model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:341e69f15a5759ae12954df8716c1830fd9455490703ebc75afaa46f58f830ac
+size 4699234800

model-00006-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:34fa27a2bebb22d7e54bcb4fbbafdbe35506d197817cd1560ac03d97ebad4468
-size 4586465768

model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddad8844e8df67f3193459b903498fe193b76dd3748922a60ae971b581344f2d
+size 4678797952

model-00007-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:be61a6358449389bdc2e4c52d9d094665ef171e4de067839f8c7f9cbca2d7830
-size 4749076088

model-00008-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d5ef0d9d001d57ca0ed408ca225d2c6ac4d9f42a2a6300f9b96e48478d5dcd63
-size 4749076088

model-00009-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6985a771159429c89c50dac303694e06a3137e08ae87ecee8aae8f9e1260a81b
-size 4586465768

model-00010-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9960ee3e641cac45264c3709fdddf23534d7905c966807f09093896301016b58
-size 4749076088

model-00011-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:aa980f1b62b8982b5d4a57f15f753775ecda395c1bfebaa5a421493ef3b5b5a4
-size 4749076088

model-00012-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:89521f2ad74fbfdb32e61dd6a3d44033700ef6e1dacb8542d0335559c1f1a49c
-size 4749076088

model-00013-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:76372da816e4ba64fe6adc503aabba5f779cd54d3c6392e1c017c239f2fc9891
-size 4586465776

model-00014-of-00014.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:df48182526d437406b2bed5cf25a7d53ad456cd23ceb65cce4beaaae6c2e1e2c
-size 3082172920

recipe.yaml

@@ -1,32 +1,6 @@
-quant_stage:
-  quant_modifiers:
+default_stage:
+  default_modifiers:
     QuantizationModifier:
-      config_groups:
-        group_0:
-          targets: [Linear]
-          weights:
-            num_bits: 8
-            type: float
-            symmetric: true
-            group_size: null
-            strategy: channel
-            block_structure: null
-            dynamic: false
-            actorder: null
-            observer: mse
-            observer_kwargs: {}
-          input_activations:
-            num_bits: 8
-            type: float
-            symmetric: true
-            group_size: null
-            strategy: token
-            block_structure: null
-            dynamic: true
-            actorder: null
-            observer: null
-            observer_kwargs: {}
-          output_activations: null
-          format: null
       targets: [Linear]
-      ignore: [lm_head, 're:.*router', 're:.*mamba']
+      ignore: [lm_head, 're:.*block_sparse_moe.router']
+      scheme: FP8_DYNAMIC