<kbd id="afajh"><form id="afajh"></form></kbd>
<strong id="afajh"><dl id="afajh"></dl></strong>
    <del id="afajh"><form id="afajh"></form></del>
        1. <th id="afajh"><progress id="afajh"></progress></th>
          <b id="afajh"><abbr id="afajh"></abbr></b>
          <th id="afajh"><progress id="afajh"></progress></th>

          手把手拆解:從零實(shí)現(xiàn)Llama3大模型(Python)

          共 25226字,需瀏覽 51分鐘

           ·

          2024-05-22 21:34


          自Meta 發(fā)布了開(kāi)源大模型 llama3 系列,在多個(gè)關(guān)鍵基準(zhǔn)測(cè)試中優(yōu)于業(yè)界 SOTA 模型,并在代碼生成任務(wù)上全面領(lǐng)先。太強(qiáng)了!10大開(kāi)源大模型!
          此后,開(kāi)發(fā)者們便開(kāi)始了本地部署和實(shí)現(xiàn),比如 llama3 的中文實(shí)現(xiàn)、llama3 的純 NumPy 實(shí)現(xiàn)等。
          近期,有位名為「Nishant Aklecha」的開(kāi)發(fā)者發(fā)布了一個(gè)從零開(kāi)始實(shí)現(xiàn) llama3 的存儲(chǔ)庫(kù),包括跨多個(gè)頭的注意力矩陣乘法、位置編碼和每個(gè)層在內(nèi)都有非常詳細(xì)的解釋。項(xiàng)目初期就已在 GitHub 上收獲了 1.5k 的 star,足可見(jiàn)其含金量!

          從零開(kāi)始實(shí)現(xiàn) llama3
          接下來(lái)項(xiàng)目作者手把手教你如何從頭開(kāi)始實(shí)現(xiàn) llama3。
          項(xiàng)目地址:https://github.com/naklecha/llama3-from-scratch
          首先從 Meta 提供的 llama3 模型文件中加載張量。
          下載地址:https://llama.meta.com/llama-downloads/
          是分詞器(tokenizer),作者表示沒(méi)打算自己實(shí)現(xiàn)分詞器,因而借用了 Andrej Karpathy 的實(shí)現(xiàn)方式:
          分詞器的實(shí)現(xiàn)鏈接:https://github.com/karpathy/minbpe
          from pathlib import Pathimport tiktokenfrom tiktoken.load import load_tiktoken_bpeimport torchimport jsonimport matplotlib.pyplot as plttokenizer_path = "Meta-Llama-3-8B/tokenizer.model"special_tokens = [            "<|begin_of_text|>",            "<|end_of_text|>",            "<|reserved_special_token_0|>",            "<|reserved_special_token_1|>",            "<|reserved_special_token_2|>",            "<|reserved_special_token_3|>",            "<|start_header_id|>",            "<|end_header_id|>",            "<|reserved_special_token_4|>",            "<|eot_id|>",  # end of turn        ] + [f"<|reserved_special_token_{i}|>" for i in range (5, 256 - 5)] mergeable_ranks = load_tiktoken_bpe (tokenizer_path) tokenizer = tiktoken.Encoding (    name=Path (tokenizer_path).name,    pat_str=r"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p {L}\p {N}]?\p {L}+|\p {N}{1,3}| ?[^\s\p {L}\p {N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+",    mergeable_ranks=mergeable_ranks,    special_tokens={token: len (mergeable_ranks) + i for i, token in enumerate (special_tokens)},)tokenizer.decode (tokenizer.encode ("hello world!"))
          'hello world!'

          上述步驟完成后,就是讀取模型文件了。由于該研究是從頭開(kāi)始實(shí)現(xiàn) llama3,因此代碼一次只讀取一個(gè)張量文件。
          model = torch.load ("Meta-Llama-3-8B/consolidated.00.pth")print (json.dumps (list (model.keys ())[:20], indent=4))
          [    "tok_embeddings.weight",    "layers.0.attention.wq.weight",    "layers.0.attention.wk.weight",    "layers.0.attention.wv.weight",    "layers.0.attention.wo.weight",    "layers.0.feed_forward.w1.weight",    "layers.0.feed_forward.w3.weight",    "layers.0.feed_forward.w2.weight",    "layers.0.attention_norm.weight",    "layers.0.ffn_norm.weight",    "layers.1.attention.wq.weight",    "layers.1.attention.wk.weight",    "layers.1.attention.wv.weight",    "layers.1.attention.wo.weight",    "layers.1.feed_forward.w1.weight",    "layers.1.feed_forward.w3.weight",    "layers.1.feed_forward.w2.weight",    "layers.1.attention_norm.weight",    "layers.1.ffn_norm.weight",    "layers.2.attention.wq.weight"]
          with open ("Meta-Llama-3-8B/params.json", "r") as f:    config = json.load (f)config
          {'dim': 4096, 'n_layers': 32, 'n_heads': 32, 'n_kv_heads': 8, 'vocab_size': 128256, 'multiple_of': 1024, 'ffn_dim_multiplier': 1.3, 'norm_eps': 1e-05, 'rope_theta': 500000.0}
          項(xiàng)目作者使用以下配置來(lái)推斷模型細(xì)節(jié):
          • 模型有 32 個(gè) transformer 層;
          • 每個(gè)多頭注意力塊有 32 個(gè)頭。


          dim = config ["dim"]n_layers = config ["n_layers"]n_heads = config ["n_heads"]n_kv_heads = config ["n_kv_heads"]vocab_size = config ["vocab_size"]multiple_of = config ["multiple_of"]ffn_dim_multiplier = config ["ffn_dim_multiplier"]norm_eps = config ["norm_eps"]rope_theta = torch.tensor (config ["rope_theta"])
          接下來(lái)的操作是將文本裝換為 token,這里作者使用的是 tiktoken 庫(kù)(一個(gè)用于 OpenAI 模型的 BPE tokeniser)。
          prompt = "the answer to the ultimate question of life, the universe, and everything is"tokens = [128000] + tokenizer.encode (prompt)print (tokens)tokens = torch.tensor (tokens)prompt_split_as_tokens = [tokenizer.decode ([token.item ()]) for token in tokens]print (prompt_split_as_tokens)
          [128000, 1820, 4320, 311, 279, 17139, 3488, 315, 2324, 11, 279, 15861, 11, 323, 4395, 374, 220]['<|begin_of_text|>', 'the', ' answer', ' to', ' the', ' ultimate', ' question', ' of', ' life', ',', ' the', ' universe', ',', ' and', ' everything', ' is', ' ']
          然后將 token 轉(zhuǎn)換為嵌入。

          embedding_layer = torch.nn.Embedding (vocab_size, dim)embedding_layer.weight.data.copy_(model ["tok_embeddings.weight"])token_embeddings_unnormalized = embedding_layer (tokens).to (torch.bfloat16)token_embeddings_unnormalized.shape
          torch.Size ([17, 4096])
          將嵌入進(jìn)行歸一化。該研究使用均方根 RMS 算法進(jìn)行歸一化。不過(guò),在這一步之后,張量形狀不會(huì)改變,只是值進(jìn)行了歸一化。

          # def rms_norm (tensor, norm_weights):#     rms = (tensor.pow (2).mean (-1, keepdim=True) + norm_eps)**0.5#     return tensor * (norm_weights /rms)def rms_norm (tensor, norm_weights):    return (tensor * torch.rsqrt (tensor.pow (2).mean (-1, keepdim=True) + norm_eps)) * norm_weights
          構(gòu)建 transformer 第一層。完成上述準(zhǔn)備后,接著是構(gòu)建 transformer 第一層:從模型文件中訪問(wèn) layer.0(即第一層),歸一化后嵌入維度仍然是 [17x4096] 。
          token_embeddings = rms_norm (token_embeddings_unnormalized, model ["layers.0.attention_norm.weight"])token_embeddings.shape
          torch.Size ([17, 4096])
          從頭開(kāi)始實(shí)現(xiàn)注意力。加載第一層 transformer 的注意力頭:
          print (    model ["layers.0.attention.wq.weight"].shape,    model ["layers.0.attention.wk.weight"].shape,    model ["layers.0.attention.wv.weight"].shape,    model ["layers.0.attention.wo.weight"].shape)torch.Size ([4096, 4096]) torch.Size ([1024, 4096]) torch.Size ([1024, 4096]) torch.Size ([4096, 4096])
          展開(kāi)查詢(xún)。展開(kāi)來(lái)自多個(gè)注意力頭的查詢(xún),得到的形狀是 [32x128x4096],這里,32 是 llama3 中注意力頭的數(shù)量,128 是查詢(xún)向量的大小,4096 是 token 嵌入的大小。
          q_layer0 = model ["layers.0.attention.wq.weight"]head_dim = q_layer0.shape [0] //n_headsq_layer0 = q_layer0.view (n_heads, head_dim, dim)q_layer0.shape
          torch.Size ([32, 128, 4096])
          從頭實(shí)現(xiàn)第一層的第一個(gè)頭。訪問(wèn)第一層的查詢(xún)權(quán)重矩陣,大小是 [128x4096]。
          q_layer0_head0 = q_layer0 [0]q_layer0_head0.shape
          torch.Size ([128, 4096])
          將查詢(xún)權(quán)重與 token 嵌入相乘,從而得到 token 的查詢(xún),在這里你可以看到結(jié)果大小是 [17x128]。
          q_per_token = torch.matmul (token_embeddings, q_layer0_head0.T)q_per_token.shape
          torch.Size ([17, 128])
          定位編碼。現(xiàn)在處于這樣一個(gè)階段,即對(duì)提示符中的每個(gè) token 都有一個(gè)查詢(xún)向量,但是考慮單個(gè)查詢(xún)向量,我們不知道其提示符中的位置。作者使用了 RoPE(旋轉(zhuǎn)位置嵌入)來(lái)解決。
          q_per_token_split_into_pairs = q_per_token.float ().view (q_per_token.shape [0], -1, 2)q_per_token_split_into_pairs.shape

          torch.Size ([17, 64, 2])

          在上面的步驟中,該研究將查詢(xún)向量分成對(duì),并對(duì)每對(duì)應(yīng)用旋轉(zhuǎn)角度移位。
          使用復(fù)數(shù)點(diǎn)積來(lái)旋轉(zhuǎn)向量。

          zero_to_one_split_into_64_parts = torch.tensor (range (64))/64zero_to_one_split_into_64_parts
          tensor ([0.0000, 0.0156, 0.0312, 0.0469, 0.0625, 0.0781, 0.0938, 0.1094, 0.1250,        0.1406, 0.1562, 0.1719, 0.1875, 0.2031, 0.2188, 0.2344, 0.2500, 0.2656,        0.2812, 0.2969, 0.3125, 0.3281, 0.3438, 0.3594, 0.3750, 0.3906, 0.4062,        0.4219, 0.4375, 0.4531, 0.4688, 0.4844, 0.5000, 0.5156, 0.5312, 0.5469,        0.5625, 0.5781, 0.5938, 0.6094, 0.6250, 0.6406, 0.6562, 0.6719, 0.6875,        0.7031, 0.7188, 0.7344, 0.7500, 0.7656, 0.7812, 0.7969, 0.8125, 0.8281,        0.8438, 0.8594, 0.8750, 0.8906, 0.9062, 0.9219, 0.9375, 0.9531, 0.9688,        0.9844])
          freqs = 1.0 / (rope_theta ** zero_to_one_split_into_64_parts)freqs
          tensor ([1.0000e+00, 8.1462e-01, 6.6360e-01, 5.4058e-01, 4.4037e-01, 3.5873e-01,        2.9223e-01, 2.3805e-01, 1.9392e-01, 1.5797e-01, 1.2869e-01, 1.0483e-01,        8.5397e-02, 6.9566e-02, 5.6670e-02, 4.6164e-02, 3.7606e-02, 3.0635e-02,        2.4955e-02, 2.0329e-02, 1.6560e-02, 1.3490e-02, 1.0990e-02, 8.9523e-03,        7.2927e-03, 5.9407e-03, 4.8394e-03, 3.9423e-03, 3.2114e-03, 2.6161e-03,        2.1311e-03, 1.7360e-03, 1.4142e-03, 1.1520e-03, 9.3847e-04, 7.6450e-04,        6.2277e-04, 5.0732e-04, 4.1327e-04, 3.3666e-04, 2.7425e-04, 2.2341e-04,        1.8199e-04, 1.4825e-04, 1.2077e-04, 9.8381e-05, 8.0143e-05, 6.5286e-05,        5.3183e-05, 4.3324e-05, 3.5292e-05, 2.8750e-05, 2.3420e-05, 1.9078e-05,        1.5542e-05, 1.2660e-05, 1.0313e-05, 8.4015e-06, 6.8440e-06, 5.5752e-06,        4.5417e-06, 3.6997e-06, 3.0139e-06, 2.4551e-06])
          freqs_for_each_token = torch.outer (torch.arange (17), freqs)freqs_cis = torch.polar (torch.ones_like (freqs_for_each_token), freqs_for_each_token)freqs_cis.shape# viewing tjhe third row of freqs_cisvalue = freqs_cis [3]plt.figure ()for i, element in enumerate (value [:17]):    plt.plot ([0, element.real], [0, element.imag], color='blue', linewidth=1, label=f"Index: {i}")    plt.annotate (f"{i}", xy=(element.real, element.imag), color='red')    plt.xlabel ('Real')    plt.ylabel ('Imaginary')    plt.title ('Plot of one row of freqs_cis')    plt.show ()

          現(xiàn)在每個(gè) token 查詢(xún)都有了復(fù)數(shù)。
          q_per_token_as_complex_numbers = torch.view_as_complex (q_per_token_split_into_pairs)q_per_token_as_complex_numbers.shape
          torch.Size ([17, 64])
          q_per_token_as_complex_numbers_rotated = q_per_token_as_complex_numbers * freqs_cisq_per_token_as_complex_numbers_rotated.shape
          torch.Size ([17, 64])

          旋轉(zhuǎn)后的向量。
          q_per_token_split_into_pairs_rotated = torch.view_as_real (q_per_token_as_complex_numbers_rotated)q_per_token_split_into_pairs_rotated.shape
          torch.Size ([17, 64, 2])

          現(xiàn)在有了一個(gè)新的查詢(xún)向量 (旋轉(zhuǎn)查詢(xún)向量),形狀為 [17x128],其中 17 是 token 數(shù)量,128 是查詢(xún)向量的維度。
          q_per_token_rotated = q_per_token_split_into_pairs_rotated.view (q_per_token.shape)q_per_token_rotated.shape
          torch.Size ([17, 128])

          鍵(幾乎和查詢(xún)一樣),鍵也生成維度為 128 的鍵向量。鍵的權(quán)重只有查詢(xún)的 1/4,這是因?yàn)殒I的權(quán)重在 4 個(gè)頭之間共享,以減少所需的計(jì)算量,鍵也會(huì)被旋轉(zhuǎn)以添加位置信息,就像查詢(xún)一樣。
          k_layer0 = model ["layers.0.attention.wk.weight"]k_layer0 = k_layer0.view (n_kv_heads, k_layer0.shape [0] //n_kv_heads, dim)k_layer0.shape
          torch.Size ([8, 128, 4096])
          k_layer0_head0 = k_layer0 [0]k_layer0_head0.shape
          torch.Size ([128, 4096])
          k_per_token = torch.matmul (token_embeddings, k_layer0_head0.T)k_per_token.shape
          torch.Size ([17, 128])
          k_per_token_split_into_pairs = k_per_token.float ().view (k_per_token.shape [0], -1, 2)k_per_token_split_into_pairs.shape
          torch.Size ([17, 64, 2])
          k_per_token_as_complex_numbers = torch.view_as_complex (k_per_token_split_into_pairs)k_per_token_as_complex_numbers.shape
          torch.Size ([17, 64])
          k_per_token_split_into_pairs_rotated = torch.view_as_real (k_per_token_as_complex_numbers * freqs_cis)k_per_token_split_into_pairs_rotated.shape
          torch.Size ([17, 64, 2])
          k_per_token_rotated = k_per_token_split_into_pairs_rotated.view (k_per_token.shape)k_per_token_rotated.shape
          torch.Size ([17, 128])

          每個(gè) token 查詢(xún)和鍵的旋轉(zhuǎn)值如下,每個(gè)查詢(xún)和鍵現(xiàn)在的形狀都是 [17x128]。
          接下來(lái)一步是將查詢(xún)和鍵矩陣相乘。注意力得分矩陣 (qk_per_token) 的形狀為 [17x17],其中 17 是提示中 token 的數(shù)量。
          qk_per_token = torch.matmul (q_per_token_rotated, k_per_token_rotated.T)/(head_dim)**0.5qk_per_token.shape
          torch.Size ([17, 17])
          現(xiàn)在必須掩蔽查詢(xún)鍵分?jǐn)?shù)。
          在 llama3 的訓(xùn)練過(guò)程中,未來(lái) token 的 qk 分?jǐn)?shù)被掩蔽。這是因?yàn)樵谟?xùn)練期間,只學(xué)習(xí)使用過(guò)去的 token 來(lái)預(yù)測(cè)未來(lái)的 token。因此在推理過(guò)程中,將未來(lái)的 token 標(biāo)記為零。
          def display_qk_heatmap (qk_per_token):    _, ax = plt.subplots ()    im = ax.imshow (qk_per_token.to (float).detach (), cmap='viridis')    ax.set_xticks (range (len (prompt_split_as_tokens)))    ax.set_yticks (range (len (prompt_split_as_tokens)))    ax.set_xticklabels (prompt_split_as_tokens)    ax.set_yticklabels (prompt_split_as_tokens)    ax.figure.colorbar (im, ax=ax)    display_qk_heatmap (qk_per_token)

          mask = torch.full ((len (tokens), len (tokens)), float ("-inf"), device=tokens.device) mask = torch.triu (mask, diagonal=1) mask
          tensor ([[0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf, -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -inf],        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]])

          qk_per_token_after_masking = qk_per_token + maskdisplay_qk_heatmap (qk_per_token_after_masking)

          qk_per_token_after_masking_after_softmax = torch.nn.functional.softmax (qk_per_token_after_masking, dim=1).to (torch.bfloat16) display_qk_heatmap (qk_per_token_after_masking_after_softmax)
          值(幾乎在注意力結(jié)束時(shí))
          這些分?jǐn)?shù) (0-1) 被用于確定每個(gè) token 使用了多少值矩陣。
          •  就像鍵一樣,值權(quán)重也在 4 個(gè)注意力頭之間共享(以節(jié)省計(jì)算量)
          •  結(jié)果,下面的值權(quán)重矩陣形狀為 [8x128x4096]
          v_layer0 = model ["layers.0.attention.wv.weight"] v_layer0 = v_layer0.view (n_kv_heads, v_layer0.shape [0] //n_kv_heads, dim) v_layer0.shape
          torch.Size ([8, 128, 4096])
          第一層和第一個(gè)頭的值權(quán)重矩陣如下所示。
          v_layer0_head0 = v_layer0 [0] v_layer0_head0.shape
          torch.Size ([128, 4096])
          值向量如下圖所示。
          現(xiàn)在使用值權(quán)重來(lái)獲取每個(gè) token 的注意力值,其大小為 [17x128],其中 17 為提示中的 token 數(shù),128 為每個(gè) token 的值向量維數(shù)。
          v_per_token = torch.matmul (token_embeddings, v_layer0_head0.T)v_per_token.shape
          torch.Size ([17, 128])
          注意力如下圖所示。
          與每個(gè) token 的值相乘后得到的注意力向量的形狀為 [17*128]。
          qkv_attention = torch.matmul (qk_per_token_after_masking_after_softmax, v_per_token) qkv_attention.shape
          torch.Size ([17, 128])
          多頭注意力與單頭注意力如下圖所示。
          現(xiàn)在有了第一層和第一個(gè)頭的注意力值。
          接下來(lái)運(yùn)行一個(gè)循環(huán)并執(zhí)行與上面單元完全相同的數(shù)學(xué)運(yùn)算,不過(guò)第一層中的每個(gè)頭除外。
          qkv_attention_store = []for head in range (n_heads):    q_layer0_head = q_layer0 [head]    k_layer0_head = k_layer0 [head//4] # key weights are shared across 4 headsv_layer0_head = v_layer0 [head//4] # value weights are shared across 4 headsq_per_token = torch.matmul (token_embeddings, q_layer0_head.T)    k_per_token = torch.matmul (token_embeddings, k_layer0_head.T)    v_per_token = torch.matmul (token_embeddings, v_layer0_head.T)

          q_per_token_split_into_pairs = q_per_token.float ().view (q_per_token.shape [0], -1, 2) q_per_token_as_complex_numbers = torch.view_as_complex (q_per_token_split_into_pairs) q_per_token_split_into_pairs_rotated = torch.view_as_real (q_per_token_as_complex_numbers * freqs_cis [:len (tokens)]) q_per_token_rotated = q_per_token_split_into_pairs_rotated.view (q_per_token.shape)

          k_per_token_split_into_pairs = k_per_token.float ().view (k_per_token.shape [0], -1, 2) k_per_token_as_complex_numbers = torch.view_as_complex (k_per_token_split_into_pairs) k_per_token_split_into_pairs_rotated = torch.view_as_real (k_per_token_as_complex_numbers * freqs_cis [:len (tokens)]) k_per_token_rotated = k_per_token_split_into_pairs_rotated.view (k_per_token.shape)

          qk_per_token = torch.matmul (q_per_token_rotated, k_per_token_rotated.T)/(128)**0.5mask = torch.full ((len (tokens), len (tokens)), float ("-inf"), device=tokens.device) mask = torch.triu (mask, diagonal=1) qk_per_token_after_masking = qk_per_token + maskqk_per_token_after_masking_after_softmax = torch.nn.functional.softmax (qk_per_token_after_masking, dim=1).to (torch.bfloat16) qkv_attention = torch.matmul (qk_per_token_after_masking_after_softmax, v_per_token) qkv_attention = torch.matmul (qk_per_token_after_masking_after_softmax, v_per_token) qkv_attention_store.append (qkv_attention)len (qkv_attention_store)
          32
          現(xiàn)在第一層上的所有 32 個(gè)頭都有了 qkv_attention 矩陣,并在快結(jié)束的時(shí)候?qū)⑺凶⒁饬Ψ謹(jǐn)?shù)合并為一個(gè)大小為 [17x4096] 的大矩陣。
          stacked_qkv_attention = torch.cat (qkv_attention_store, dim=-1) stacked_qkv_attention.shape
          torch.Size ([17, 4096])
          權(quán)重矩陣是最后的步驟之一。
          第 0 層注意力要做的最后一件事是,對(duì)以下的權(quán)重矩陣進(jìn)行乘法操作。
          w_layer0 = model ["layers.0.attention.wo.weight"] w_layer0.shape
          torch.Size ([4096, 4096])
          這是一個(gè)簡(jiǎn)單的線(xiàn)性層,所以只做矩陣乘法(matmul)。
          embedding_delta = torch.matmul (stacked_qkv_attention, w_layer0.T) embedding_delta.shape
          torch.Size ([17, 4096])
          現(xiàn)在,注意力之后的嵌入值有了變化,并應(yīng)該被添加到原始 token 嵌入中。
          embedding_after_edit = token_embeddings_unnormalized + embedding_deltaembedding_after_edit.shape
          torch.Size ([17, 4096])
          歸一化并在嵌入 delta 過(guò)程中運(yùn)行一個(gè)前饋神經(jīng)網(wǎng)絡(luò)。
          embedding_after_edit_normalized = rms_norm (embedding_after_edit, model ["layers.0.ffn_norm.weight"]) embedding_after_edit_normalized.shape
          torch.Size ([17, 4096])
          加載 ff 權(quán)重,并實(shí)現(xiàn)前饋網(wǎng)絡(luò)。

          llama3 使用 SwiGLU前饋網(wǎng)絡(luò),該網(wǎng)絡(luò)架構(gòu)非常擅長(zhǎng)在模型需要時(shí)添加非線(xiàn)性。當(dāng)前,在 LLMs 中使用這一前饋網(wǎng)絡(luò)是非常標(biāo)準(zhǔn)的做法。
          w1 = model ["layers.0.feed_forward.w1.weight"] w2 = model ["layers.0.feed_forward.w2.weight"] w3 = model ["layers.0.feed_forward.w3.weight"] output_after_feedforward = torch.matmul (torch.functional.F.silu (torch.matmul (embedding_after_edit_normalized, w1.T)) * torch.matmul (embedding_after_edit_normalized, w3.T), w2.T) output_after_feedforward.shape
          torch.Size ([17, 4096])
          現(xiàn)在終于在第一層之后為每個(gè) token 提供了新的編輯后的嵌入,并且在完成之前只剩下 31 層需要處理(one for loop away)。
          你可以想象這個(gè)編輯后的嵌入擁有在第一層上所有查詢(xún)的信息。現(xiàn)在每一層將在所問(wèn)問(wèn)題上編碼越來(lái)越復(fù)雜的查詢(xún),直到得到的嵌入了解所需的下一個(gè) token 的一切。
          layer_0_embedding = embedding_after_edit+output_after_feedforwardlayer_0_embedding.shape
          torch.Size ([17, 4096])
          之前為每一層做的所有事情,都可以一次性完成。
          final_embedding = token_embeddings_unnormalizedfor layer in range (n_layers):    qkv_attention_store = []    layer_embedding_norm = rms_norm (final_embedding, model [f"layers.{layer}.attention_norm.weight"])    q_layer = model [f"layers.{layer}.attention.wq.weight"]    q_layer = q_layer.view (n_heads, q_layer.shape [0] //n_heads, dim)    k_layer = model [f"layers.{layer}.attention.wk.weight"]    k_layer = k_layer.view (n_kv_heads, k_layer.shape [0] //n_kv_heads, dim)    v_layer = model [f"layers.{layer}.attention.wv.weight"]    v_layer = v_layer.view (n_kv_heads, v_layer.shape [0] //n_kv_heads, dim)    w_layer = model [f"layers.{layer}.attention.wo.weight"]    for head in range (n_heads):        q_layer_head = q_layer [head]        k_layer_head = k_layer [head//4]        v_layer_head = v_layer [head//4]        q_per_token = torch.matmul (layer_embedding_norm, q_layer_head.T)        k_per_token = torch.matmul (layer_embedding_norm, k_layer_head.T)        v_per_token = torch.matmul (layer_embedding_norm, v_layer_head.T)        q_per_token_split_into_pairs = q_per_token.float ().view (q_per_token.shape [0], -1, 2)        q_per_token_as_complex_numbers = torch.view_as_complex (q_per_token_split_into_pairs)        q_per_token_split_into_pairs_rotated = torch.view_as_real (q_per_token_as_complex_numbers * freqs_cis)        q_per_token_rotated = q_per_token_split_into_pairs_rotated.view (q_per_token.shape)        k_per_token_split_into_pairs = k_per_token.float ().view (k_per_token.shape [0], -1, 2)        k_per_token_as_complex_numbers = torch.view_as_complex (k_per_token_split_into_pairs)        k_per_token_split_into_pairs_rotated = torch.view_as_real (k_per_token_as_complex_numbers * freqs_cis)        k_per_token_rotated = k_per_token_split_into_pairs_rotated.view (k_per_token.shape)        qk_per_token = torch.matmul (q_per_token_rotated, k_per_token_rotated.T)/(128)**0.5        mask = torch.full ((len (token_embeddings_unnormalized), len (token_embeddings_unnormalized)), float ("-inf"))        mask = torch.triu (mask, diagonal=1)        qk_per_token_after_masking = qk_per_token + mask        qk_per_token_after_masking_after_softmax = torch.nn.functional.softmax (qk_per_token_after_masking, dim=1).to (torch.bfloat16)        qkv_attention = torch.matmul (qk_per_token_after_masking_after_softmax, v_per_token)        qkv_attention_store.append (qkv_attention)

          stacked_qkv_attention = torch.cat (qkv_attention_store, dim=-1) w_layer = model [f"layers.{layer}.attention.wo.weight"] embedding_delta = torch.matmul (stacked_qkv_attention, w_layer.T) embedding_after_edit = final_embedding + embedding_delta embedding_after_edit_normalized = rms_norm (embedding_after_edit, model [f"layers.{layer}.ffn_norm.weight"]) w1 = model [f"layers.{layer}.feed_forward.w1.weight"] w2 = model [f"layers.{layer}.feed_forward.w2.weight"] w3 = model [f"layers.{layer}.feed_forward.w3.weight"] output_after_feedforward = torch.matmul (torch.functional.F.silu (torch.matmul (embedding_after_edit_normalized, w1.T)) * torch.matmul (embedding_after_edit_normalized, w3.T), w2.T) final_embedding = embedding_after_edit+output_after_feedforward
          現(xiàn)在有了最終的嵌入,即該模型對(duì)下一個(gè) token 的最佳猜測(cè)。該嵌入的形狀與常見(jiàn)的 token 嵌入 [17x4096] 相同,其中 17 為 token 數(shù),4096 為嵌入維數(shù)。
          final_embedding = rms_norm (final_embedding, model ["norm.weight"]) final_embedding.shape
          torch.Size ([17, 4096])
          將該嵌入解碼為 token 值。
          使用該輸入解碼器將最終的嵌入轉(zhuǎn)換為一個(gè) token。
          model ["output.weight"].shape
          torch.Size ([128256, 4096])
          使用最后 token 的嵌入來(lái)預(yù)測(cè)下一個(gè)值。在示例中,42 是「生命、宇宙和萬(wàn)物終極問(wèn)題的答案是什么」的答案,根據(jù)《銀河系漫游指南》一書(shū),大多數(shù)現(xiàn)代 LLMs 都會(huì)回答 42,應(yīng)該驗(yàn)證了整個(gè)代碼。
          logits = torch.matmul (final_embedding [-1], model ["output.weight"].T) logits.shape
          torch.Size ([128256])
          模型預(yù)測(cè) token 數(shù) 2983 為下一個(gè) token,這是 42 的 token 數(shù)嗎?以下是最后的代碼單元。
          next_token = torch.argmax (logits, dim=-1) next_token
          tensor (2983)
          最后,啟動(dòng)。

          tokenizer.decode ([next_token.item ()])
          '42'

          結(jié)~


          瀏覽 158
          2點(diǎn)贊
          評(píng)論
          收藏
          分享

          手機(jī)掃一掃分享

          分享
          舉報(bào)
          評(píng)論
          圖片
          表情
          推薦
          2點(diǎn)贊
          評(píng)論
          收藏
          分享

          手機(jī)掃一掃分享

          分享
          舉報(bào)
          <kbd id="afajh"><form id="afajh"></form></kbd>
          <strong id="afajh"><dl id="afajh"></dl></strong>
            <del id="afajh"><form id="afajh"></form></del>
                1. <th id="afajh"><progress id="afajh"></progress></th>
                  <b id="afajh"><abbr id="afajh"></abbr></b>
                  <th id="afajh"><progress id="afajh"></progress></th>
                  台湾中文成人在线 | 国产乱╳╳╳╳性视频大全 | 成人精品电影久久 | 日本A视频在线观看 | 婷婷综合网 性 |