New references

yandexdataschool · Dec 3, 2023 · e2b2ebc · e2b2ebc
1 parent eb76283
commit e2b2ebc
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 19 deletions.
diff --git a/week10_efficiency/gptq_block_error_reference.pt b/week10_efficiency/gptq_block_error_reference.pt
diff --git a/week10_efficiency/gptq_block_weight_reference.pt b/week10_efficiency/gptq_block_weight_reference.pt
diff --git a/week10_efficiency/gptq_weight_reference.pt b/week10_efficiency/gptq_weight_reference.pt
diff --git a/week10_efficiency/practice.ipynb b/week10_efficiency/practice.ipynb
@@ -122,7 +122,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 4,
+      "execution_count": 3,
       "metadata": {
         "id": "z1xDv2d12hw9"
       },
@@ -182,7 +182,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 6,
+      "execution_count": 5,
       "metadata": {
         "id": "9ZR0OZ0ai62N"
       },
@@ -271,7 +271,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 8,
+      "execution_count": 7,
       "metadata": {
         "id": "NMzGxzt-i62P"
       },
@@ -331,7 +331,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 9,
+      "execution_count": null,
       "metadata": {
         "id": "JIUbMToO3C9B"
       },
@@ -856,7 +856,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 11,
+      "execution_count": 10,
       "metadata": {
         "id": "ls5meN1c4No-"
       },
@@ -931,7 +931,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 12,
+      "execution_count": 11,
       "metadata": {
         "id": "8vo0UYKvi62V"
       },
@@ -973,7 +973,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 13,
+      "execution_count": 12,
       "metadata": {
         "id": "2TRwggs3Lb2F"
       },
@@ -1055,7 +1055,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 15,
+      "execution_count": 14,
       "metadata": {
         "id": "luUgnOhxaFcw"
       },
@@ -1153,7 +1153,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": null,
+      "execution_count": 17,
       "metadata": {
         "colab": {
           "base_uri": "https://localhost:8080/"
@@ -1195,7 +1195,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": null,
+      "execution_count": 18,
       "metadata": {
         "id": "Fn-Vko3urrM0"
       },
@@ -1383,7 +1383,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 18,
+      "execution_count": 21,
       "metadata": {
         "id": "7W50rdH0i62Y"
       },
@@ -1449,7 +1449,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 20,
+      "execution_count": 23,
       "metadata": {
         "id": "zzEPMSsEi62Y"
       },
@@ -1626,7 +1626,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": null,
+      "execution_count": 26,
       "metadata": {
         "id": "urb_7OhSq_zA"
       },
@@ -1723,7 +1723,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 21,
+      "execution_count": 45,
       "metadata": {
         "id": "t8GwQU5CrNv_"
       },
@@ -1781,7 +1781,7 @@
         "block_weight = weight[:,:16]\n",
         "\n",
         "block_hessian_inverse = (torch.triu(torch.rand((16, 16), generator=generator), diagonal=1) + torch.diag(torch.rand(16, generator=generator) + 1)).cuda()\n",
-        "quantized_block_weight, scaled_block_error = gptq_block(block_weight, block_hessian_inverse, scale, zero, 15)\n",
+        "quantized_block_weight, scaled_block_error = gptq_block(block_weight, block_hessian_inverse, scale, zero, 4)\n",
         "\n",
         "assert torch.all(quantized_block_weight == torch.load(\"gptq_block_weight_reference.pt\"))\n",
         "assert torch.allclose(scaled_block_error, torch.load(\"gptq_block_error_reference.pt\"), rtol=1e-5, atol=1e-06)\n",
@@ -1812,7 +1812,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 25,
+      "execution_count": 47,
       "metadata": {
         "id": "iqI4FamprFPE"
       },
@@ -1911,7 +1911,7 @@
         "\n",
         "hessian = torch.triu(torch.rand((128, 128), generator=generator) + 4 * torch.eye(128)).cuda().half()\n",
         "hessian += hessian.clone().T\n",
-        "quantized_weight, _, _ = gptq(weight, 8, hessian, 32)\n",
+        "quantized_weight, _, _ = gptq(weight, 4, hessian, 32)\n",
         "\n",
         "assert torch.all(quantized_weight == torch.load(\"gptq_weight_reference.pt\"))\n",
         "\n",
@@ -1944,7 +1944,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 27,
+      "execution_count": 49,
       "metadata": {
         "id": "fMyrGkHqy-xg"
       },
@@ -2103,7 +2103,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": null,
+      "execution_count": 53,
       "metadata": {
         "id": "18BHd4Nnza2T"
       },