From e2ea9cf9a3bfa509d5ad8c011c23469a9e66f6d4 Mon Sep 17 00:00:00 2001 From: Noam Gat Date: Thu, 19 Oct 2023 13:20:06 +0300 Subject: [PATCH] Added llama.cpp integration notebook --- .../colab_llamacpppython_integration.ipynb | 772 ++++++++++++++++++ 1 file changed, 772 insertions(+) create mode 100644 samples/colab_llamacpppython_integration.ipynb diff --git a/samples/colab_llamacpppython_integration.ipynb b/samples/colab_llamacpppython_integration.ipynb new file mode 100644 index 0000000..02d99bf --- /dev/null +++ b/samples/colab_llamacpppython_integration.ipynb @@ -0,0 +1,772 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# LM Format Enforcer Integration with llama.cpp (python bindings)\n", + "\n", + "\n", + " \"Open\n", + "\n", + "\n", + "This notebook shows how you can integrate with the llama.cpp library via its [python bindings](https://github.com/abetlen/llama-cpp-python). We will do this using its ```LogitsProcessor``` interface, and show how we integrate with ~30 lines of code for the connection.\n", + "\n", + "This sample notebook focuses on simplicity and ease of setup. Therefore we will use a CPU version of llamacpp, which will make inference slower. For production use, you should use the GPU version of llamacpp." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Installing dependencies\n", + "\n", + "We begin by installing the dependencies.\n", + "\n" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [], + "source": [ + "!pip install llama-cpp-python lm-format-enforcer huggingface-hub\n", + "\n", + "# When running from source / developing the library, use this instead\n", + "# %load_ext autoreload\n", + "# %autoreload 2\n", + "# import sys\n", + "# import os\n", + "# sys.path.append(os.path.abspath('..'))\n", + "## os.environ['CUDA_LAUNCH_BLOCKING'] = '1'" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Loading the model\n", + "\n", + "This demo uses [Llama2 gguf weights by TheBloke](https://huggingface.co/TheBloke/Llama-2-7b-Chat-GGUF). We will use huggingface hub to download the model." + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from /mnt/e/manual/llama-2-7b-chat.Q5_K_M.gguf (version GGUF V2 (latest))\n", + "llama_model_loader: - tensor 0: token_embd.weight q5_K [ 4096, 32000, 1, 1 ]\n", + "llama_model_loader: - tensor 1: blk.0.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 2: blk.0.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 3: blk.0.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 4: blk.0.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 5: blk.0.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 6: blk.0.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 7: blk.0.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 8: blk.0.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 9: blk.0.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 10: blk.1.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 11: blk.1.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 12: blk.1.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 13: blk.1.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 14: blk.1.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 15: blk.1.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 16: blk.1.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 17: blk.1.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 18: blk.1.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 19: blk.10.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 20: blk.10.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 21: blk.10.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 22: blk.10.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 23: blk.10.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 24: blk.10.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 25: blk.10.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 26: blk.10.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 27: blk.10.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 28: blk.11.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 29: blk.11.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 30: blk.11.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 31: blk.11.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 32: blk.11.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 33: blk.11.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 34: blk.11.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 35: blk.11.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 36: blk.11.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 37: blk.12.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 38: blk.12.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 39: blk.12.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 40: blk.12.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 41: blk.12.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 42: blk.12.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 43: blk.12.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 44: blk.12.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 45: blk.12.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 46: blk.13.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 47: blk.13.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 48: blk.13.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 49: blk.13.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 50: blk.13.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 51: blk.13.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 52: blk.13.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 53: blk.13.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 54: blk.13.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 55: blk.14.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 56: blk.14.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 57: blk.14.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 58: blk.14.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 59: blk.14.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 60: blk.14.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 61: blk.14.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 62: blk.14.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 63: blk.14.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 64: blk.15.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 65: blk.15.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 66: blk.15.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 67: blk.15.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 68: blk.15.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 69: blk.15.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 70: blk.15.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 71: blk.15.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 72: blk.15.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 73: blk.16.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 74: blk.16.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 75: blk.16.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 76: blk.16.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 77: blk.16.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 78: blk.16.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 79: blk.16.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 80: blk.16.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 81: blk.16.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 82: blk.17.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 83: blk.17.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 84: blk.17.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 85: blk.17.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 86: blk.17.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 87: blk.17.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 88: blk.17.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 89: blk.17.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 90: blk.17.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 91: blk.18.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 92: blk.18.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 93: blk.18.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 94: blk.18.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 95: blk.18.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 96: blk.18.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 97: blk.18.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 98: blk.18.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 99: blk.18.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 100: blk.19.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 101: blk.19.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 102: blk.19.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 103: blk.19.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 104: blk.19.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 105: blk.19.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 106: blk.19.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 107: blk.19.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 108: blk.19.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 109: blk.2.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 110: blk.2.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 111: blk.2.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 112: blk.2.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 113: blk.2.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 114: blk.2.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 115: blk.2.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 116: blk.2.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 117: blk.2.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 118: blk.20.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 119: blk.20.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 120: blk.20.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 121: blk.20.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 122: blk.20.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 123: blk.20.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 124: blk.20.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 125: blk.20.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 126: blk.20.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 127: blk.21.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 128: blk.21.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 129: blk.21.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 130: blk.21.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 131: blk.21.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 132: blk.21.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 133: blk.21.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 134: blk.21.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 135: blk.21.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 136: blk.22.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 137: blk.22.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 138: blk.22.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 139: blk.22.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 140: blk.22.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 141: blk.22.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 142: blk.22.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 143: blk.22.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 144: blk.22.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 145: blk.23.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 146: blk.23.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 147: blk.23.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 148: blk.23.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 149: blk.23.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 150: blk.23.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 151: blk.23.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 152: blk.23.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 153: blk.23.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 154: blk.3.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 155: blk.3.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 156: blk.3.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 157: blk.3.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 158: blk.3.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 159: blk.3.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 160: blk.3.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 161: blk.3.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 162: blk.3.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 163: blk.4.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 164: blk.4.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 165: blk.4.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 166: blk.4.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 167: blk.4.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 168: blk.4.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 169: blk.4.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 170: blk.4.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 171: blk.4.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 172: blk.5.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 173: blk.5.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 174: blk.5.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 175: blk.5.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 176: blk.5.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 177: blk.5.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 178: blk.5.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 179: blk.5.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 180: blk.5.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 181: blk.6.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 182: blk.6.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 183: blk.6.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 184: blk.6.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 185: blk.6.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 186: blk.6.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 187: blk.6.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 188: blk.6.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 189: blk.6.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 190: blk.7.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 191: blk.7.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 192: blk.7.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 193: blk.7.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 194: blk.7.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 195: blk.7.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 196: blk.7.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 197: blk.7.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 198: blk.7.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 199: blk.8.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 200: blk.8.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 201: blk.8.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 202: blk.8.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 203: blk.8.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 204: blk.8.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 205: blk.8.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 206: blk.8.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 207: blk.8.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 208: blk.9.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 209: blk.9.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 210: blk.9.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 211: blk.9.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 212: blk.9.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 213: blk.9.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 214: blk.9.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 215: blk.9.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 216: blk.9.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 217: output.weight q6_K [ 4096, 32000, 1, 1 ]\n", + "llama_model_loader: - tensor 218: blk.24.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 219: blk.24.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 220: blk.24.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 221: blk.24.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 222: blk.24.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 223: blk.24.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 224: blk.24.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 225: blk.24.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 226: blk.24.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 227: blk.25.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 228: blk.25.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 229: blk.25.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 230: blk.25.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 231: blk.25.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 232: blk.25.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 233: blk.25.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 234: blk.25.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 235: blk.25.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 236: blk.26.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 237: blk.26.ffn_down.weight q5_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 238: blk.26.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 239: blk.26.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 240: blk.26.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 241: blk.26.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 242: blk.26.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 243: blk.26.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 244: blk.26.attn_v.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 245: blk.27.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 246: blk.27.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 247: blk.27.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 248: blk.27.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 249: blk.27.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 250: blk.27.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 251: blk.27.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 252: blk.27.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 253: blk.27.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 254: blk.28.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 255: blk.28.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 256: blk.28.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 257: blk.28.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 258: blk.28.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 259: blk.28.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 260: blk.28.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 261: blk.28.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 262: blk.28.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 263: blk.29.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 264: blk.29.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 265: blk.29.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 266: blk.29.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 267: blk.29.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 268: blk.29.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 269: blk.29.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 270: blk.29.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 271: blk.29.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 272: blk.30.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 273: blk.30.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 274: blk.30.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 275: blk.30.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 276: blk.30.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 277: blk.30.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 278: blk.30.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 279: blk.30.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 280: blk.30.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 281: blk.31.attn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 282: blk.31.ffn_down.weight q6_K [ 11008, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 283: blk.31.ffn_gate.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 284: blk.31.ffn_up.weight q5_K [ 4096, 11008, 1, 1 ]\n", + "llama_model_loader: - tensor 285: blk.31.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - tensor 286: blk.31.attn_k.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 287: blk.31.attn_output.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 288: blk.31.attn_q.weight q5_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 289: blk.31.attn_v.weight q6_K [ 4096, 4096, 1, 1 ]\n", + "llama_model_loader: - tensor 290: output_norm.weight f32 [ 4096, 1, 1, 1 ]\n", + "llama_model_loader: - kv 0: general.architecture str \n", + "llama_model_loader: - kv 1: general.name str \n", + "llama_model_loader: - kv 2: llama.context_length u32 \n", + "llama_model_loader: - kv 3: llama.embedding_length u32 \n", + "llama_model_loader: - kv 4: llama.block_count u32 \n", + "llama_model_loader: - kv 5: llama.feed_forward_length u32 \n", + "llama_model_loader: - kv 6: llama.rope.dimension_count u32 \n", + "llama_model_loader: - kv 7: llama.attention.head_count u32 \n", + "llama_model_loader: - kv 8: llama.attention.head_count_kv u32 \n", + "llama_model_loader: - kv 9: llama.attention.layer_norm_rms_epsilon f32 \n", + "llama_model_loader: - kv 10: general.file_type u32 \n", + "llama_model_loader: - kv 11: tokenizer.ggml.model str \n", + "llama_model_loader: - kv 12: tokenizer.ggml.tokens arr \n", + "llama_model_loader: - kv 13: tokenizer.ggml.scores arr \n", + "llama_model_loader: - kv 14: tokenizer.ggml.token_type arr \n", + "llama_model_loader: - kv 15: tokenizer.ggml.bos_token_id u32 \n", + "llama_model_loader: - kv 16: tokenizer.ggml.eos_token_id u32 \n", + "llama_model_loader: - kv 17: tokenizer.ggml.unknown_token_id u32 \n", + "llama_model_loader: - kv 18: general.quantization_version u32 \n", + "llama_model_loader: - type f32: 65 tensors\n", + "llama_model_loader: - type q5_K: 193 tensors\n", + "llama_model_loader: - type q6_K: 33 tensors\n", + "llm_load_print_meta: format = GGUF V2 (latest)\n", + "llm_load_print_meta: arch = llama\n", + "llm_load_print_meta: vocab type = SPM\n", + "llm_load_print_meta: n_vocab = 32000\n", + "llm_load_print_meta: n_merges = 0\n", + "llm_load_print_meta: n_ctx_train = 4096\n", + "llm_load_print_meta: n_embd = 4096\n", + "llm_load_print_meta: n_head = 32\n", + "llm_load_print_meta: n_head_kv = 32\n", + "llm_load_print_meta: n_layer = 32\n", + "llm_load_print_meta: n_rot = 128\n", + "llm_load_print_meta: n_gqa = 1\n", + "llm_load_print_meta: f_norm_eps = 0.0e+00\n", + "llm_load_print_meta: f_norm_rms_eps = 1.0e-06\n", + "llm_load_print_meta: n_ff = 11008\n", + "llm_load_print_meta: freq_base_train = 10000.0\n", + "llm_load_print_meta: freq_scale_train = 1\n", + "llm_load_print_meta: model type = 7B\n", + "llm_load_print_meta: model ftype = mostly Q5_K - Medium\n", + "llm_load_print_meta: model params = 6.74 B\n", + "llm_load_print_meta: model size = 4.45 GiB (5.68 BPW) \n", + "llm_load_print_meta: general.name = LLaMA v2\n", + "llm_load_print_meta: BOS token = 1 ''\n", + "llm_load_print_meta: EOS token = 2 ''\n", + "llm_load_print_meta: UNK token = 0 ''\n", + "llm_load_print_meta: LF token = 13 '<0x0A>'\n", + "llm_load_tensors: ggml ctx size = 0.09 MB\n", + "llm_load_tensors: mem required = 4560.96 MB\n", + "...................................................................................................\n", + "llama_new_context_with_model: n_ctx = 512\n", + "llama_new_context_with_model: freq_base = 10000.0\n", + "llama_new_context_with_model: freq_scale = 1\n", + "llama_new_context_with_model: kv self size = 256.00 MB\n", + "llama_new_context_with_model: compute buffer total size = 76.38 MB\n", + "AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | \n" + ] + } + ], + "source": [ + "from llama_cpp import Llama\n", + "from huggingface_hub import hf_hub_download\n", + "downloaded_model_path = hf_hub_download(repo_id=\"TheBloke/Llama-2-7b-Chat-GGUF\", filename=\"llama-2-7b-chat.Q5_K_M.gguf\")\n", + "llm = Llama(model_path=downloaded_model_path)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "If the previous cell executed successfully, you have propertly set up your Colab runtime and loaded the llama.cpp model!" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Creating a Logits Processor that filters tokens\n", + "\n", + "llama.cpp's python bindigs have a ```LogitsProcessor``` interface similar to one that exists in Huggingface Transformers. We will connect to this API and set the logits that are not allowed to negative infinity, ensuring they are not selected.\n", + "\n", + "We use the high level llama.cpp python interface to create a ```TokenEnforcer```, and a ```LogitsProcessor``` that uses it." + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "WARNING:root:Could not import transformerenforcer. Transformers-based functionality will not be available. Details: No module named 'transformers'\n" + ] + } + ], + "source": [ + "from llama_cpp import LogitsProcessor, LogitsProcessorList\n", + "from lmformatenforcer import CharacterLevelParser, TokenEnforcer\n", + "import numpy as np\n", + "import numpy.typing as npt\n", + "from typing import Tuple, List\n", + "\n", + "def _build_regular_tokens_list(llm: Llama) -> List[Tuple[int, str]]:\n", + " token_0 = llm.tokenize(b\"0\")[-1]\n", + " regular_tokens = []\n", + " special_tokens = [llm.token_bos(), llm.token_eos()]\n", + " for token_idx in range(llm.n_vocab()):\n", + " if token_idx in special_tokens:\n", + " continue\n", + " # We prepend token 0 and skip the first letter of the result to get a space if the token is a start word.\n", + " try:\n", + " decoded = llm.detokenize([token_0, token_idx]).decode('utf-8')[1:]\n", + " regular_tokens.append((token_idx, decoded))\n", + " except:\n", + " # This can happen for cases such as raw bytes outside of the ASCII range. We ignore them and never allow them.\n", + " pass\n", + " return regular_tokens\n", + "\n", + "\n", + "def build_llamacpp_logits_processor(llm: Llama, character_level_parser: CharacterLevelParser) -> LogitsProcessor:\n", + " \"\"\"Build the logits processor function that llama.cpp will use to filter the tokens generated by the model. The result\n", + " can be passed in the logits_processor list that is sent to the call or generate() method of llama.cpp models.\"\"\"\n", + " regular_tokens = _build_regular_tokens_list(llm)\n", + " def decoder(sent: List[int]) -> str:\n", + " return llm.detokenize(sent).decode('utf-8')\n", + " token_enforcer = TokenEnforcer(regular_tokens, character_level_parser, decoder, llm.token_eos())\n", + "\n", + " def llamacpp_logits_processor(input_ids: npt.NDArray[np.intc], scores: npt.NDArray[np.single]) -> npt.NDArray[np.single]:\n", + " token_sequence = input_ids.tolist()\n", + " allowed_tokens = token_enforcer.get_allowed_tokens(token_sequence)\n", + " mask = np.ones(scores.shape, bool)\n", + " mask[allowed_tokens] = False\n", + " scores[mask] = float('-inf')\n", + " return scores\n", + " \n", + " return llamacpp_logits_processor\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "A few helper functions to make display nicer." + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "metadata": {}, + "outputs": [], + "source": [ + "from IPython.display import display, Markdown\n", + "\n", + "def display_header(text):\n", + " display(Markdown(f'**{text}**'))\n", + "\n", + "def display_content(text):\n", + " display(Markdown(f'```\\n{text}\\n```'))" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Setting up the prompt for the specific language model\n", + "\n", + "We set up the prompting style according to the [Llama2 demo](https://huggingface.co/spaces/huggingface-projects/llama-2-13b-chat/blob/main/app.py). We simplify the implementation a bit as we don't need chat history for this demo." + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "metadata": {}, + "outputs": [], + "source": [ + "DEFAULT_SYSTEM_PROMPT = \"\"\"\\\n", + "You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\\n\\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\\\n", + "\"\"\"\n", + "\n", + "def get_prompt(message: str, system_prompt: str = DEFAULT_SYSTEM_PROMPT) -> str:\n", + " return f'[INST] <>\\n{system_prompt}\\n<>\\n\\n{message} [/INST]'" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Generating text with the LM Format Enforcer Logits Processor\n", + "In order to integrate our logits processor with LlamaCpp, we create a ```LogitsProcessorList``` and pass it as a keyword variable when using the ```Llama``` class.\n" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "metadata": {}, + "outputs": [], + "source": [ + "from typing import Optional\n", + "\n", + "def llamacpp_with_character_level_parser(llm: Llama, prompt: str, character_level_parser: Optional[CharacterLevelParser]) -> str:\n", + " logits_processors: Optional[LogitsProcessorList] = None\n", + " if character_level_parser:\n", + " logits_processors = LogitsProcessorList([build_llamacpp_logits_processor(llm, character_level_parser)])\n", + " \n", + " output = llm(prompt, logits_processor=logits_processors)\n", + " text: str = output['choices'][0]['text']\n", + " return text" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## LlamaCpp + JSON Use case\n", + "\n", + "Now we demonstrate using ```JsonSchemaParser```. We create a pydantic model, generate the schema from it, and use that to enforce the format.\n", + "The output will always be in a format that can be parsed by the parser." + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/tmp/ipykernel_401088/4169945469.py:13: PydanticDeprecatedSince20: The `schema_json` method is deprecated; use `model_json_schema` and json.dumps instead. Deprecated in Pydantic V2.0 to be removed in V3.0. See Pydantic V2 Migration Guide at https://errors.pydantic.dev/2.4/migration/\n", + " question_with_schema = f'{question}{AnswerFormat.schema_json()}'\n" + ] + }, + { + "data": { + "text/markdown": [ + "**Prompt:**" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/markdown": [ + "```\n", + "[INST] <>\n", + "You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n", + "\n", + "If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\n", + "<>\n", + "\n", + "Please give me information about Michael Jordan. You MUST answer using the following json schema: {\"properties\": {\"first_name\": {\"title\": \"First Name\", \"type\": \"string\"}, \"last_name\": {\"title\": \"Last Name\", \"type\": \"string\"}, \"year_of_birth\": {\"title\": \"Year Of Birth\", \"type\": \"integer\"}, \"num_seasons_in_nba\": {\"title\": \"Num Seasons In Nba\", \"type\": \"integer\"}}, \"required\": [\"first_name\", \"last_name\", \"year_of_birth\", \"num_seasons_in_nba\"], \"title\": \"AnswerFormat\", \"type\": \"object\"} [/INST]\n", + "```" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/markdown": [ + "**Answer, With json schema enforcing:**" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/tmp/ipykernel_401088/4169945469.py:21: PydanticDeprecatedSince20: The `schema` method is deprecated; use `model_json_schema` instead. Deprecated in Pydantic V2.0 to be removed in V3.0. See Pydantic V2 Migration Guide at https://errors.pydantic.dev/2.4/migration/\n", + " result = llamacpp_with_character_level_parser(llm, prompt, JsonSchemaParser(AnswerFormat.schema()))\n", + "\n", + "llama_print_timings: load time = 16791.06 ms\n", + "llama_print_timings: sample time = 17.44 ms / 53 runs ( 0.33 ms per token, 3038.64 tokens per second)\n", + "llama_print_timings: prompt eval time = 16791.00 ms / 294 tokens ( 57.11 ms per token, 17.51 tokens per second)\n", + "llama_print_timings: eval time = 4649.13 ms / 52 runs ( 89.41 ms per token, 11.18 tokens per second)\n", + "llama_print_timings: total time = 21662.33 ms\n" + ] + }, + { + "data": { + "text/markdown": [ + "```\n", + " { \"first_name\": \"Michael\", \"last_name\": \"Jordan\", \"year_of_birth\": 1963, \"num_seasons_in_nba\": 15 }\n", + "\n", + "\n", + "\n", + "```" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/markdown": [ + "**Answer, Without json schema enforcing:**" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Llama.generate: prefix-match hit\n", + "\n", + "llama_print_timings: load time = 16791.06 ms\n", + "llama_print_timings: sample time = 34.38 ms / 99 runs ( 0.35 ms per token, 2879.58 tokens per second)\n", + "llama_print_timings: prompt eval time = 0.00 ms / 1 tokens ( 0.00 ms per token, inf tokens per second)\n", + "llama_print_timings: eval time = 8800.92 ms / 99 runs ( 88.90 ms per token, 11.25 tokens per second)\n", + "llama_print_timings: total time = 8956.60 ms\n" + ] + }, + { + "data": { + "text/markdown": [ + "```\n", + " Of course! I'd be happy to provide information about Michael Jordan using the provided JSON schema. Here is the information for you:\n", + "{\n", + "\"first_name\": \"Michael\",\n", + "\"last_name\": \"Jordan\",\n", + "\"year_of_birth\": 1963,\n", + "\"num_seasons_in_nba\": 15\n", + "}\n", + "I hope this helps! Let me know if you have any other questions.\n", + "```" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "from lmformatenforcer import JsonSchemaParser\n", + "from pydantic import BaseModel\n", + "\n", + "from typing import List\n", + "\n", + "class AnswerFormat(BaseModel):\n", + " first_name: str\n", + " last_name: str\n", + " year_of_birth: int\n", + " num_seasons_in_nba: int\n", + "\n", + "question = 'Please give me information about Michael Jordan. You MUST answer using the following json schema: '\n", + "question_with_schema = f'{question}{AnswerFormat.schema_json()}'\n", + "prompt = get_prompt(question_with_schema)\n", + "\n", + "display_header(\"Prompt:\")\n", + "display_content(prompt)\n", + "\n", + "display_header(\"Answer, With json schema enforcing:\")\n", + "\n", + "result = llamacpp_with_character_level_parser(llm, prompt, JsonSchemaParser(AnswerFormat.schema()))\n", + "display_content(result)\n", + "\n", + "display_header(\"Answer, Without json schema enforcing:\")\n", + "result = llamacpp_with_character_level_parser(llm, prompt, None)\n", + "display_content(result)\n", + "\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "As you can see, the enforced output matches the required schema, while the unenforced does not. We have successfully integrated with llama.cpp!" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "lmformatenforcer", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.10.12" + }, + "orig_nbformat": 4 + }, + "nbformat": 4, + "nbformat_minor": 2 +}