transformers/pr_43265/en/_app/immutable/nodes/153.1cd2cb1f.js

import{s as Po,o as Do,n as S}from"../chunks/scheduler.31fdf58d.js";import{S as Ko,i as Oo,e as d,s,c as g,h as en,a as c,d as r,b as a,f as X,j as f,g as _,k as q,w as tn,l,m as i,n as y,t as M,o as T,p as b}from"../chunks/index.2f76fdf0.js";import{T as ho}from"../chunks/Tip.8d349121.js";import{C as on}from"../chunks/CopyLLMTxtMenu.53b607bf.js";import{D as ne}from"../chunks/Docstring.7acc6835.js";import{C as E}from"../chunks/CodeBlock.e52df5d6.js";import{E as nt}from"../chunks/ExampleCodeBlock.f9704f52.js";import{H as et,E as nn}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.08750ec0.js";import{H as sn,a as uo}from"../chunks/HfOption.fb051768.js";function an(J){let t,p;return t=new E({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMHBpcGVsaW5lJTBBJTBBJTBBcGlwZWxpbmUlMjAlM0QlMjBwaXBlbGluZSglMEElMjAlMjAlMjAlMjB0YXNrJTNEJTIydGV4dC1nZW5lcmF0aW9uJTIyJTJDJTBBJTIwJTIwJTIwJTIwbW9kZWwlM0QlMjJDb2hlcmVGb3JBSSUyRmM0YWktY29tbWFuZC1yLXYwMSUyMiUyQyUwQSUyMCUyMCUyMCUyMGRldmljZSUzRDAlMEEpJTBBcGlwZWxpbmUoJTIyUGxhbnRzJTIwY3JlYXRlJTIwZW5lcmd5JTIwdGhyb3VnaCUyMGElMjBwcm9jZXNzJTIwa25vd24lMjBhcyUyMik=",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline


pipeline = pipeline(
    task=<span class="hljs-string">&quot;text-generation&quot;</span>,
    model=<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>,
    device=<span class="hljs-number">0</span>
)
pipeline(<span class="hljs-string">&quot;Plants create energy through a process known as&quot;</span>)`,lang:"python",wrap:!1}}),{c(){g(t.$$.fragment)},l(o){_(t.$$.fragment,o)},m(o,h){y(t,o,h),p=!0},p:S,i(o){p||(M(t.$$.fragment,o),p=!0)},o(o){T(t.$$.fragment,o),p=!1},d(o){b(t,o)}}}function rn(J){let t,p;return t=new E({props:{code:"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",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer


tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>, device_map=<span class="hljs-string">&quot;auto&quot;</span>, attn_implementation=<span class="hljs-string">&quot;sdpa&quot;</span>)

<span class="hljs-comment"># format message with the Command-R chat template</span>
messages = [{<span class="hljs-string">&quot;role&quot;</span>: <span class="hljs-string">&quot;user&quot;</span>, <span class="hljs-string">&quot;content&quot;</span>: <span class="hljs-string">&quot;How do plants make energy?&quot;</span>}]
input_ids = tokenizer.apply_chat_template(messages, tokenize=<span class="hljs-literal">True</span>, add_generation_prompt=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>).to(model.device)
output = model.generate(
    input_ids,
    max_new_tokens=<span class="hljs-number">100</span>,
    do_sample=<span class="hljs-literal">True</span>,
    temperature=<span class="hljs-number">0.3</span>,
    cache_implementation=<span class="hljs-string">&quot;static&quot;</span>,
)
<span class="hljs-built_in">print</span>(tokenizer.decode(output[<span class="hljs-number">0</span>], skip_special_tokens=<span class="hljs-literal">True</span>))`,lang:"python",wrap:!1}}),{c(){g(t.$$.fragment)},l(o){_(t.$$.fragment,o)},m(o,h){y(t,o,h),p=!0},p:S,i(o){p||(M(t.$$.fragment,o),p=!0)},o(o){T(t.$$.fragment,o),p=!1},d(o){b(t,o)}}}function ln(J){let t,p;return t=new E({props:{code:"JTIzJTIwcGlwJTIwaW5zdGFsbCUyMC1VJTIwZmxhc2gtYXR0biUyMC0tbm8tYnVpbGQtaXNvbGF0aW9uJTBBdHJhbnNmb3JtZXJzJTIwY2hhdCUyMENvaGVyZUZvckFJJTJGYzRhaS1jb21tYW5kLXItdjAxJTIwLS1kdHlwZSUyMGF1dG8lMjAtLWF0dG5faW1wbGVtZW50YXRpb24lMjBmbGFzaF9hdHRlbnRpb25fMg==",highlighted:`<span class="hljs-comment"># pip install -U flash-attn --no-build-isolation</span>
transformers chat CohereForAI/c4ai-command-r-v01 --dtype auto --attn_implementation flash_attention_2`,lang:"bash",wrap:!1}}),{c(){g(t.$$.fragment)},l(o){_(t.$$.fragment,o)},m(o,h){y(t,o,h),p=!0},p:S,i(o){p||(M(t.$$.fragment,o),p=!0)},o(o){T(t.$$.fragment,o),p=!1},d(o){b(t,o)}}}function dn(J){let t,p,o,h,w,m;return t=new uo({props:{id:"usage",option:"Pipeline",$$slots:{default:[an]},$$scope:{ctx:J}}}),o=new uo({props:{id:"usage",option:"AutoModel",$$slots:{default:[rn]},$$scope:{ctx:J}}}),w=new uo({props:{id:"usage",option:"transformers CLI",$$slots:{default:[ln]},$$scope:{ctx:J}}}),{c(){g(t.$$.fragment),p=s(),g(o.$$.fragment),h=s(),g(w.$$.fragment)},l(u){_(t.$$.fragment,u),p=a(u),_(o.$$.fragment,u),h=a(u),_(w.$$.fragment,u)},m(u,U){y(t,u,U),i(u,p,U),y(o,u,U),i(u,h,U),y(w,u,U),m=!0},p(u,U){const se={};U&2&&(se.$$scope={dirty:U,ctx:u}),t.$set(se);const B={};U&2&&(B.$$scope={dirty:U,ctx:u}),o.$set(B);const tt={};U&2&&(tt.$$scope={dirty:U,ctx:u}),w.$set(tt)},i(u){m||(M(t.$$.fragment,u),M(o.$$.fragment,u),M(w.$$.fragment,u),m=!0)},o(u){T(t.$$.fragment,u),T(o.$$.fragment,u),T(w.$$.fragment,u),m=!1},d(u){u&&(r(p),r(h)),b(t,u),b(o,u),b(w,u)}}}function cn(J){let t,p;return t=new E({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMENvaGVyZU1vZGVsJTJDJTIwQ29oZXJlQ29uZmlnJTBBJTBBJTIzJTIwSW5pdGlhbGl6aW5nJTIwYSUyMENvaGVyZSUyMG1vZGVsJTIwY29uZmlndXJhdGlvbiUwQWNvbmZpZ3VyYXRpb24lMjAlM0QlMjBDb2hlcmVDb25maWcoKSUwQSUwQSUyMyUyMEluaXRpYWxpemluZyUyMGElMjBtb2RlbCUyMGZyb20lMjB0aGUlMjBDb2hlcmUlMjBjb25maWd1cmF0aW9uJTBBbW9kZWwlMjAlM0QlMjBDb2hlcmVNb2RlbChjb25maWd1cmF0aW9uKSUwQSUyMyUyMEFjY2Vzc2luZyUyMHRoZSUyMG1vZGVsJTIwY29uZmlndXJhdGlvbiUwQWNvbmZpZ3VyYXRpb24lMjAlM0QlMjBtb2RlbC5jb25maWc=",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> CohereModel, CohereConfig

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a Cohere model configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = CohereConfig()

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a model from the Cohere configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = CohereModel(configuration)
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Accessing the model configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = model.config`,lang:"python",wrap:!1}}),{c(){g(t.$$.fragment)},l(o){_(t.$$.fragment,o)},m(o,h){y(t,o,h),p=!0},p:S,i(o){p||(M(t.$$.fragment,o),p=!0)},o(o){T(t.$$.fragment,o),p=!1},d(o){b(t,o)}}}function pn(J){let t,p;return t=new E({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Ub2tlbml6ZXIlMEElMEF0b2tlbml6ZXIlMjAlM0QlMjBBdXRvVG9rZW5pemVyLmZyb21fcHJldHJhaW5lZCglMjJDb2hlcmVGb3JBSSUyRmM0YWktY29tbWFuZC1yLXYwMSUyMiklMEF0b2tlbml6ZXIuZW5jb2RlKCUyMkhlbGxvJTIwdGhpcyUyMGlzJTIwYSUyMHRlc3QlMjIp",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer.encode(<span class="hljs-string">&quot;Hello this is a test&quot;</span>)
[<span class="hljs-number">5</span>, <span class="hljs-number">28339</span>, <span class="hljs-number">2075</span>, <span class="hljs-number">1801</span>, <span class="hljs-number">1671</span>, <span class="hljs-number">3282</span>]`,lang:"python",wrap:!1}}),{c(){g(t.$$.fragment)},l(o){_(t.$$.fragment,o)},m(o,h){y(t,o,h),p=!0},p:S,i(o){p||(M(t.$$.fragment,o),p=!0)},o(o){T(t.$$.fragment,o),p=!1},d(o){b(t,o)}}}function mn(J){let t,p="When used with <code>is_split_into_words=True</code>, this tokenizer needs to be instantiated with <code>add_prefix_space=True</code>.";return{c(){t=d("p"),t.innerHTML=p},l(o){t=c(o,"P",{"data-svelte-h":!0}),f(t)!=="svelte-9gg91e"&&(t.innerHTML=p)},m(o,h){i(o,t,h)},p:S,d(o){o&&r(t)}}}function hn(J){let t,p="Examples:",o,h,w;return h=new E({props:{code:"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",highlighted:`&gt;&gt; tokenizer = CohereTokenizer.from_pretrained(<span class="hljs-string">&#x27;CohereForAI/c4ai-command-r-v01&#x27;</span>)

&gt;&gt; <span class="hljs-comment"># define documents:</span>
&gt;&gt; documents = [
    { <span class="hljs-string">&quot;title&quot;</span>: <span class="hljs-string">&quot;Tall penguins&quot;</span>, <span class="hljs-string">&quot;text&quot;</span>: <span class="hljs-string">&quot;Emperor penguins are the tallest.&quot;</span> },
    { <span class="hljs-string">&quot;title&quot;</span>: <span class="hljs-string">&quot;Penguin habitats&quot;</span>, <span class="hljs-string">&quot;text&quot;</span>: <span class="hljs-string">&quot;Emperor penguins only live in Antarctica.&quot;</span>}
]
&gt;&gt; <span class="hljs-comment"># define a conversation:</span>
&gt;&gt; conversation = [
    {<span class="hljs-string">&quot;role&quot;</span>: <span class="hljs-string">&quot;user&quot;</span>, <span class="hljs-string">&quot;content&quot;</span>: <span class="hljs-string">&quot;Whats the biggest penguin in the world?&quot;</span>}
]
&gt;&gt; <span class="hljs-comment"># render the prompt, ready for user to inspect, or for input into the model:</span>
&gt;&gt; grounded_generation_prompt = tokenizer.apply_grounded_generation_template(conversation, documents=documents, tokenize=<span class="hljs-literal">False</span>, add_generation_prompt=<span class="hljs-literal">True</span>)
&gt;&gt; <span class="hljs-built_in">print</span>(grounded_generation_prompt)
&gt;&gt; inputs = tokenizer.encode(prompt, add_special_tokens=<span class="hljs-literal">False</span>, return_tensors=<span class="hljs-string">&#x27;pt&#x27;</span>)
&gt;&gt; outputs = model.generate(inputs, max_new_tokens=<span class="hljs-number">128</span>)
&gt;&gt; <span class="hljs-built_in">print</span>(tokenizer.decode(outputs[<span class="hljs-number">0</span>]))`,lang:"python",wrap:!1}}),{c(){t=d("p"),t.textContent=p,o=s(),g(h.$$.fragment)},l(m){t=c(m,"P",{"data-svelte-h":!0}),f(t)!=="svelte-kvfsh7"&&(t.textContent=p),o=a(m),_(h.$$.fragment,m)},m(m,u){i(m,t,u),i(m,o,u),y(h,m,u),w=!0},p:S,i(m){w||(M(h.$$.fragment,m),w=!0)},o(m){T(h.$$.fragment,m),w=!1},d(m){m&&(r(t),r(o)),b(h,m)}}}function un(J){let t,p="Examples:",o,h,w;return h=new E({props:{code:"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",highlighted:`&gt;&gt; tokenizer = CohereTokenizer.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
&gt;&gt; tools = [
    {
        <span class="hljs-string">&quot;name&quot;</span>: <span class="hljs-string">&quot;internet_search&quot;</span>,
        <span class="hljs-string">&quot;description&quot;</span>: <span class="hljs-string">&quot;Returns a list of relevant document snippets for a textual query retrieved from the internet&quot;</span>,
        <span class="hljs-string">&quot;parameter_definitions&quot;</span>: {
            <span class="hljs-string">&quot;query&quot;</span>: {
                <span class="hljs-string">&quot;description&quot;</span>: <span class="hljs-string">&quot;Query to search the internet with&quot;</span>,
                <span class="hljs-string">&quot;type&quot;</span>: <span class="hljs-string">&quot;str&quot;</span>,
                <span class="hljs-string">&quot;required&quot;</span>: <span class="hljs-literal">True</span>,
            }
        },
    },
    {
        <span class="hljs-string">&quot;name&quot;</span>: <span class="hljs-string">&quot;directly_answer&quot;</span>,
        <span class="hljs-string">&quot;description&quot;</span>: <span class="hljs-string">&quot;Calls a standard (un-augmented) AI chatbot to generate a response given the conversation history&quot;</span>,
        <span class="hljs-string">&quot;parameter_definitions&quot;</span>: {},
    },
]
&gt;&gt; conversation = [
    {<span class="hljs-string">&quot;role&quot;</span>: <span class="hljs-string">&quot;user&quot;</span>, <span class="hljs-string">&quot;content&quot;</span>: <span class="hljs-string">&quot;Whats the biggest penguin in the world?&quot;</span>},
]
&gt;&gt; <span class="hljs-comment"># Render the prompt, ready for user to inspect, or for input into the model</span>
&gt;&gt; prompt = tokenizer.apply_tool_use_template(conversation, tools=tools, tokenize=<span class="hljs-literal">False</span>, add_generation_prompt=<span class="hljs-literal">True</span>)
&gt;&gt; <span class="hljs-built_in">print</span>(prompt)
&gt;&gt; inputs = tokenizer.encode(grounded_generation_prompt, add_special_tokens=<span class="hljs-literal">False</span>, return_tensors=<span class="hljs-string">&#x27;pt&#x27;</span>)
&gt;&gt; outputs = model.generate(inputs, max_new_tokens=<span class="hljs-number">128</span>)
&gt;&gt; <span class="hljs-built_in">print</span>(tokenizer.decode(outputs[<span class="hljs-number">0</span>]))
[
    {
        <span class="hljs-string">&quot;tool_name&quot;</span>: <span class="hljs-string">&quot;internet_search&quot;</span>,
        <span class="hljs-string">&quot;parameters&quot;</span>: {
            <span class="hljs-string">&quot;query&quot;</span>: <span class="hljs-string">&quot;biggest penguin in the world&quot;</span>
        }
    }
]`,lang:"python",wrap:!1}}),{c(){t=d("p"),t.textContent=p,o=s(),g(h.$$.fragment)},l(m){t=c(m,"P",{"data-svelte-h":!0}),f(t)!=="svelte-kvfsh7"&&(t.textContent=p),o=a(m),_(h.$$.fragment,m)},m(m,u){i(m,t,u),i(m,o,u),y(h,m,u),w=!0},p:S,i(m){w||(M(h.$$.fragment,m),w=!0)},o(m){T(h.$$.fragment,m),w=!1},d(m){m&&(r(t),r(o)),b(h,m)}}}function fn(J){let t,p=`Although the recipe for forward pass needs to be defined within this function, one should call the <code>Module</code>
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.`;return{c(){t=d("p"),t.innerHTML=p},l(o){t=c(o,"P",{"data-svelte-h":!0}),f(t)!=="svelte-fincs2"&&(t.innerHTML=p)},m(o,h){i(o,t,h)},p:S,d(o){o&&r(t)}}}function gn(J){let t,p=`Although the recipe for forward pass needs to be defined within this function, one should call the <code>Module</code>
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.`;return{c(){t=d("p"),t.innerHTML=p},l(o){t=c(o,"P",{"data-svelte-h":!0}),f(t)!=="svelte-fincs2"&&(t.innerHTML=p)},m(o,h){i(o,t,h)},p:S,d(o){o&&r(t)}}}function _n(J){let t,p="Example:",o,h,w;return h=new E({props:{code:"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",highlighted:`&gt;&gt; <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, CohereForCausalLM

&gt;&gt; model = CohereForCausalLM.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
&gt;&gt; tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)

&gt;&gt; prompt = <span class="hljs-string">&quot;Hey, are you conscious? Can you talk to me?&quot;</span>
&gt;&gt; inputs = tokenizer(prompt, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)

&gt;&gt; <span class="hljs-comment"># Generate</span>
&gt;&gt; generate_ids = model.generate(inputs.input_ids, max_length=<span class="hljs-number">30</span>)
&gt;&gt; tokenizer.batch_decode(generate_ids, skip_special_tokens=<span class="hljs-literal">True</span>, clean_up_tokenization_spaces=<span class="hljs-literal">False</span>)[<span class="hljs-number">0</span>]
<span class="hljs-string">&quot;Hey, are you conscious? Can you talk to me?\\nI&#x27;m not conscious, but I can talk to you.&quot;</span>`,lang:"python",wrap:!1}}),{c(){t=d("p"),t.textContent=p,o=s(),g(h.$$.fragment)},l(m){t=c(m,"P",{"data-svelte-h":!0}),f(t)!=="svelte-11lpom8"&&(t.textContent=p),o=a(m),_(h.$$.fragment,m)},m(m,u){i(m,t,u),i(m,o,u),y(h,m,u),w=!0},p:S,i(m){w||(M(h.$$.fragment,m),w=!0)},o(m){T(h.$$.fragment,m),w=!1},d(m){m&&(r(t),r(o)),b(h,m)}}}function yn(J){let t,p,o,h,w,m="<em>This model was contributed to Hugging Face Transformers on 2024-03-15.</em>",u,U,se,B,tt='<div class="flex flex-wrap space-x-1"><img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat"/> <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&amp;logo=pytorch&amp;logoColor=white"/> <img alt="Tensor parallelism" src="https://img.shields.io/badge/Tensor%20parallelism-06b6d4?style=flat&amp;logoColor=white"/></div>',st,ae,at,re,fo='Cohere <a href="https://cohere.com/blog/command-r" rel="nofollow">Command-R</a> is a 35B parameter multilingual large language model designed for long context tasks like retrieval-augmented generation (RAG) and calling external APIs and tools. The model is specifically trained for grounded generation and supports both single-step and multi-step tool use. It supports a context length of 128K tokens.',rt,le,go='You can find all the original Command-R checkpoints under the <a href="https://huggingface.co/collections/CohereForAI/command-models-67652b401665205e17b192ad" rel="nofollow">Command Models</a> collection.',lt,A,_o="<p>Click on the Cohere models in the right sidebar for more examples of how to apply Cohere to different language tasks.</p>",it,ie,yo='The example below demonstrates how to generate text with <a href="/docs/transformers/pr_43265/en/main_classes/pipelines#transformers.Pipeline">Pipeline</a> or the <a href="/docs/transformers/pr_43265/en/model_doc/auto#transformers.AutoModel">AutoModel</a>, and from the command line.',dt,Q,ct,de,Mo='Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the <a href="../quantization/overview">Quantization</a> overview for more available quantization backends.',pt,ce,To='The example below uses <a href="../quantization/bitsandbytes">bitsandbytes</a> to quantize the weights to 4-bits.',mt,pe,ht,me,bo='Use the <a href="https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139" rel="nofollow">AttentionMaskVisualizer</a> to better understand what tokens the model can and cannot attend to.',ut,he,ft,L,wo='<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/cohere-attn-mask.png"/>',gt,ue,_t,fe,Jo='<li>Don’t use the dtype parameter in <a href="/docs/transformers/pr_43265/en/model_doc/auto#transformers.AutoModel.from_pretrained">from_pretrained()</a> if you’re using FlashAttention-2 because it only supports fp16 or bf16. You should use <a href="https://pytorch.org/tutorials/recipes/recipes/amp_recipe.html" rel="nofollow">Automatic Mixed Precision</a>, set fp16 or bf16 to True if using <a href="/docs/transformers/pr_43265/en/main_classes/trainer#transformers.Trainer">Trainer</a>, or use <a href="https://pytorch.org/docs/stable/amp.html#torch.autocast" rel="nofollow">torch.autocast</a>.</li>',yt,ge,Mt,G,_e,It,Ie,Co=`This is the configuration class to store the configuration of a CohereModel. It is used to instantiate a Cohere
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
defaults will yield a similar configuration to that of the <a href="https://huggingface.co/CohereForAI/c4ai-command-r-v01" rel="nofollow">CohereForAI/c4ai-command-r-v01</a>`,$t,$e,ko=`Configuration objects inherit from <a href="/docs/transformers/pr_43265/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> and can be used to control the model outputs. Read the
documentation from <a href="/docs/transformers/pr_43265/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> for more information.`,zt,Y,Tt,ye,bt,C,Me,xt,ze,Uo="Construct a Cohere tokenizer. Based on byte-level Byte-Pair-Encoding.",Zt,xe,vo="This uses notably ByteFallback and NFC normalization.",Gt,P,Ft,Ze,jo=`If you want to change the <code>bos_token</code> or the <code>eos_token</code>, make sure to specify them when initializing the model, or
call <code>tokenizer.update_post_processor()</code> to make sure that the post-processing is correctly done (otherwise the
values of the first token and final token of an encoded sequence will not be correct). For more details, checkout
[post-processors] (<a href="https://huggingface.co/docs/tokenizers/api/post-processors" rel="nofollow">https://huggingface.co/docs/tokenizers/api/post-processors</a>) documentation.`,Wt,Ge,Io=`You can get around that behavior by passing <code>add_prefix_space=True</code> when instantiating this tokenizer, but since
the model was not pretrained this way, it might yield a decrease in performance.`,Nt,D,Rt,Fe,$o=`This tokenizer inherits from <a href="/docs/transformers/pr_43265/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a> which contains most of the main methods. Users should
refer to this superclass for more information regarding those methods.`,qt,v,Te,Bt,We,zo="Create a Command-R grounded generation (aka RAG) prompt.",Vt,Ne,xo="Once rendered, the prompt instructs the model to generate a response with citations in, based on supplied documents.",Ht,Re,Zo=`Conceptually, this works in the same way as <code>apply_chat_format</code>, but takes additional <code>documents</code>
and parameter <code>citation_mode</code> parameters.`,St,qe,Go=`Converts a list of dictionaries with <code>&quot;role&quot;</code> and <code>&quot;content&quot;</code> keys and a list of
documents for the model to ground its response on into a prompt string, or a list of token ids.
This method will use the tokenizer’s <code>grounded_generation_template</code> template specified at the class level.
You can override the default template using the <code>grounded_generation_template</code> kwarg but the quality of your results may decrease.`,Xt,K,Et,j,be,At,Be,Fo="Create a Command-R tool-use prompt.",Qt,Ve,Wo=`Once rendered, the prompt instructs the model to generate a list of actions to perform on a set of user supplied tools
to help carry out the user’s requests.`,Lt,He,No="Conceptually, this works in the same way as <code>apply_chat_format</code>, but takes an additional <code>tools</code> parameter.",Yt,Se,Ro=`Converts a chat in the form of a list of dictionaries with <code>&quot;role&quot;</code> and <code>&quot;content&quot;</code> keys and a list of available
tools for the model to use into a prompt string, or a list of token ids.
This method will use the tokenizer’s <code>default_tool_use_template</code> template specified at the class level.
You can override the default template using the <code>tool_use_template</code> kwarg but the quality of your results may decrease.`,Pt,O,wt,we,Jt,I,Je,Dt,Xe,qo="The bare Cohere Model outputting raw hidden-states without any specific head on top.",Kt,Ee,Bo=`This model inherits from <a href="/docs/transformers/pr_43265/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,Ot,Ae,Vo=`This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.`,eo,R,Ce,to,Qe,Ho='The <a href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereModel">CohereModel</a> forward method, overrides the <code>__call__</code> special method.',oo,ee,no,Le,So=`<li><p><strong>last_hidden_state</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>) — Sequence of hidden-states at the output of the last layer of the model.</p> <p>If <code>past_key_values</code> is used only the last hidden-state of the sequences of shape <code>(batch_size, 1, hidden_size)</code> is output.</p></li> <li><p><strong>past_key_values</strong> (<code>Cache</code>, <em>optional</em>, returned when <code>use_cache=True</code> is passed or when <code>config.use_cache=True</code>) — It is a <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.Cache">Cache</a> instance. For more details, see our <a href="https://huggingface.co/docs/transformers/en/kv_cache" rel="nofollow">kv cache guide</a>.</p> <p>Contains pre-computed hidden-states (key and values in the self-attention blocks and optionally if
<code>config.is_encoder_decoder=True</code> in the cross-attention blocks) that can be used (see <code>past_key_values</code>
input) to speed up sequential decoding.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li>`,Ct,ke,kt,$,Ue,so,Ye,Xo="The Cohere Model for causal language modeling.",ao,Pe,Eo=`This model inherits from <a href="/docs/transformers/pr_43265/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,ro,De,Ao=`This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.`,lo,Z,ve,io,Ke,Qo='The <a href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereForCausalLM">CohereForCausalLM</a> forward method, overrides the <code>__call__</code> special method.',co,te,po,Oe,Lo=`<li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Language modeling loss (for next-token prediction).</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.vocab_size)</code>) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).</p></li> <li><p><strong>past_key_values</strong> (<code>Cache</code>, <em>optional</em>, returned when <code>use_cache=True</code> is passed or when <code>config.use_cache=True</code>) — It is a <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.Cache">Cache</a> instance. For more details, see our <a href="https://huggingface.co/docs/transformers/en/kv_cache" rel="nofollow">kv cache guide</a>.</p> <p>Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
<code>past_key_values</code> input) to speed up sequential decoding.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li>`,mo,oe,Ut,je,vt,ot,jt;return U=new on({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),ae=new et({props:{title:"Cohere",local:"cohere",headingTag:"h1"}}),Q=new sn({props:{id:"usage",options:["Pipeline","AutoModel","transformers CLI"],$$slots:{default:[dn]},$$scope:{ctx:J}}}),pe=new E({props:{code:"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",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig


bnb_config = BitsAndBytesConfig(load_in_4bit=<span class="hljs-literal">True</span>)
tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>, device_map=<span class="hljs-string">&quot;auto&quot;</span>, quantization_config=bnb_config, attn_implementation=<span class="hljs-string">&quot;sdpa&quot;</span>)

<span class="hljs-comment"># format message with the Command-R chat template</span>
messages = [{<span class="hljs-string">&quot;role&quot;</span>: <span class="hljs-string">&quot;user&quot;</span>, <span class="hljs-string">&quot;content&quot;</span>: <span class="hljs-string">&quot;How do plants make energy?&quot;</span>}]
input_ids = tokenizer.apply_chat_template(messages, tokenize=<span class="hljs-literal">True</span>, add_generation_prompt=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>).to(model.device)
output = model.generate(
    input_ids,
    max_new_tokens=<span class="hljs-number">100</span>,
    do_sample=<span class="hljs-literal">True</span>,
    temperature=<span class="hljs-number">0.3</span>,
    cache_implementation=<span class="hljs-string">&quot;static&quot;</span>,
)
<span class="hljs-built_in">print</span>(tokenizer.decode(output[<span class="hljs-number">0</span>], skip_special_tokens=<span class="hljs-literal">True</span>))`,lang:"python",wrap:!1}}),he=new E({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycy51dGlscy5hdHRlbnRpb25fdmlzdWFsaXplciUyMGltcG9ydCUyMEF0dGVudGlvbk1hc2tWaXN1YWxpemVyJTBBJTBBJTBBdmlzdWFsaXplciUyMCUzRCUyMEF0dGVudGlvbk1hc2tWaXN1YWxpemVyKCUyMkNvaGVyZUZvckFJJTJGYzRhaS1jb21tYW5kLXItdjAxJTIyKSUwQXZpc3VhbGl6ZXIoJTIyUGxhbnRzJTIwY3JlYXRlJTIwZW5lcmd5JTIwdGhyb3VnaCUyMGElMjBwcm9jZXNzJTIwa25vd24lMjBhcyUyMik=",highlighted:`<span class="hljs-keyword">from</span> transformers.utils.attention_visualizer <span class="hljs-keyword">import</span> AttentionMaskVisualizer


visualizer = AttentionMaskVisualizer(<span class="hljs-string">&quot;CohereForAI/c4ai-command-r-v01&quot;</span>)
visualizer(<span class="hljs-string">&quot;Plants create energy through a process known as&quot;</span>)`,lang:"python",wrap:!1}}),ue=new et({props:{title:"Notes",local:"notes",headingTag:"h2"}}),ge=new et({props:{title:"CohereConfig",local:"transformers.CohereConfig",headingTag:"h2"}}),_e=new ne({props:{name:"class transformers.CohereConfig",anchor:"transformers.CohereConfig",parameters:[{name:"transformers_version",val:": str | None = None"},{name:"architectures",val:": list[str] | None = None"},{name:"output_hidden_states",val:": bool | None = False"},{name:"return_dict",val:": bool | None = True"},{name:"dtype",val:": typing.Union[str, ForwardRef('torch.dtype'), NoneType] = None"},{name:"chunk_size_feed_forward",val:": int = 0"},{name:"is_encoder_decoder",val:": bool = False"},{name:"id2label",val:": dict[int, str] | dict[str, str] | None = None"},{name:"label2id",val:": dict[str, int] | dict[str, str] | None = None"},{name:"problem_type",val:": typing.Optional[typing.Literal['regression', 'single_label_classification', 'multi_label_classification']] = None"},{name:"vocab_size",val:": int = 256000"},{name:"hidden_size",val:": int = 8192"},{name:"intermediate_size",val:": int = 22528"},{name:"logit_scale",val:": float | None = 0.0625"},{name:"num_hidden_layers",val:": int = 40"},{name:"num_attention_heads",val:": int = 64"},{name:"num_key_value_heads",val:": int | None = None"},{name:"hidden_act",val:": str = 'silu'"},{name:"max_position_embeddings",val:": int = 8192"},{name:"initializer_range",val:": float = 0.02"},{name:"layer_norm_eps",val:": float | None = 1e-05"},{name:"use_cache",val:": bool = True"},{name:"pad_token_id",val:": int | None = 0"},{name:"bos_token_id",val:": int | None = 5"},{name:"eos_token_id",val:": int | list[int] | None = 255001"},{name:"tie_word_embeddings",val:": bool = True"},{name:"rope_parameters",val:": transformers.modeling_rope_utils.RopeParameters | dict | None = None"},{name:"attention_bias",val:": bool = False"},{name:"attention_dropout",val:": float | int | None = 0.0"},{name:"use_qk_norm",val:": bool | None = False"}],parametersDescription:[{anchor:"transformers.CohereConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>256000</code>) &#x2014;
Vocabulary size of the model. Defines the number of different tokens that can be represented by the <code>input_ids</code>.`,name:"vocab_size"},{anchor:"transformers.CohereConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>8192</code>) &#x2014;
Dimension of the hidden representations.`,name:"hidden_size"},{anchor:"transformers.CohereConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>22528</code>) &#x2014;
Dimension of the MLP representations.`,name:"intermediate_size"},{anchor:"transformers.CohereConfig.logit_scale",description:`<strong>logit_scale</strong> (<code>float</code>, <em>optional</em>, defaults to 0.0625) &#x2014;
The scaling factor for the output logits.`,name:"logit_scale"},{anchor:"transformers.CohereConfig.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to <code>40</code>) &#x2014;
Number of hidden layers in the Transformer decoder.`,name:"num_hidden_layers"},{anchor:"transformers.CohereConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to <code>64</code>) &#x2014;
Number of attention heads for each attention layer in the Transformer decoder.`,name:"num_attention_heads"},{anchor:"transformers.CohereConfig.num_key_value_heads",description:`<strong>num_key_value_heads</strong> (<code>int</code>, <em>optional</em>) &#x2014;
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
<code>num_key_value_heads=num_attention_heads</code>, the model will use Multi Head Attention (MHA), if
<code>num_key_value_heads=1</code> the model will use Multi Query Attention (MQA) otherwise GQA is used. When
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
by meanpooling all the original heads within that group. For more details, check out <a href="https://huggingface.co/papers/2305.13245" rel="nofollow">this
paper</a>. If it is not specified, will default to
<code>num_attention_heads</code>.`,name:"num_key_value_heads"},{anchor:"transformers.CohereConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code>, <em>optional</em>, defaults to <code>silu</code>) &#x2014;
The non-linear activation function (function or string) in the decoder. For example, <code>&quot;gelu&quot;</code>,
<code>&quot;relu&quot;</code>, <code>&quot;silu&quot;</code>, etc.`,name:"hidden_act"},{anchor:"transformers.CohereConfig.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to <code>8192</code>) &#x2014;
The maximum sequence length that this model might ever be used with.`,name:"max_position_embeddings"},{anchor:"transformers.CohereConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0.02</code>) &#x2014;
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.CohereConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1e-05</code>) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.CohereConfig.use_cache",description:`<strong>use_cache</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether or not the model should return the last key/values attentions (not used by all models). Only
relevant if <code>config.is_decoder=True</code> or when the model is a decoder-only generative model.`,name:"use_cache"},{anchor:"transformers.CohereConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to <code>0</code>) &#x2014;
Token id used for padding in the vocabulary.`,name:"pad_token_id"},{anchor:"transformers.CohereConfig.bos_token_id",description:`<strong>bos_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to <code>5</code>) &#x2014;
Token id used for beginning-of-stream in the vocabulary.`,name:"bos_token_id"},{anchor:"transformers.CohereConfig.eos_token_id",description:`<strong>eos_token_id</strong> (<code>Union[int, list[int]]</code>, <em>optional</em>, defaults to <code>255001</code>) &#x2014;
Token id used for end-of-stream in the vocabulary.`,name:"eos_token_id"},{anchor:"transformers.CohereConfig.tie_word_embeddings",description:`<strong>tie_word_embeddings</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to tie weight embeddings according to model&#x2019;s <code>tied_weights_keys</code> mapping.`,name:"tie_word_embeddings"},{anchor:"transformers.CohereConfig.rope_parameters",description:`<strong>rope_parameters</strong> (<code>Union[~modeling_rope_utils.RopeParameters, dict]</code>, <em>optional</em>) &#x2014;
Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain
a value for <code>rope_theta</code> and optionally parameters used for scaling in case you want to use RoPE
with longer <code>max_position_embeddings</code>.`,name:"rope_parameters"},{anchor:"transformers.CohereConfig.attention_bias",description:`<strong>attention_bias</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to use a bias in the query, key, value and output projection layers during self-attention.`,name:"attention_bias"},{anchor:"transformers.CohereConfig.attention_dropout",description:`<strong>attention_dropout</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.0</code>) &#x2014;
The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.CohereConfig.use_qk_norm",description:`<strong>use_qk_norm</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to use query-key normalization in the attention.`,name:"use_qk_norm"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/configuration_cohere.py#L30"}}),Y=new nt({props:{anchor:"transformers.CohereConfig.example",$$slots:{default:[cn]},$$scope:{ctx:J}}}),ye=new et({props:{title:"CohereTokenizer",local:"transformers.CohereTokenizer",headingTag:"h2"}}),Me=new ne({props:{name:"class transformers.CohereTokenizer",anchor:"transformers.CohereTokenizer",parameters:[{name:"vocab",val:": str | dict[str, int] | None = None"},{name:"merges",val:": str | list[str] | None = None"},{name:"errors",val:": str = 'replace'"},{name:"unk_token",val:": str = '<UNK>'"},{name:"bos_token",val:": str = '<BOS_TOKEN>'"},{name:"eos_token",val:": str = '<|END_OF_TURN_TOKEN|>'"},{name:"pad_token",val:": str = '<PAD>'"},{name:"cls_token",val:": str = '<CLS>'"},{name:"sep_token",val:": str = '<SEP>'"},{name:"mask_token",val:": str = '<MASK_TOKEN>'"},{name:"use_default_system_prompt",val:": bool = False"},{name:"add_prefix_space",val:": bool = False"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CohereTokenizer.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the vocabulary file.`,name:"vocab_file"},{anchor:"transformers.CohereTokenizer.merges_file",description:`<strong>merges_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the merges file.`,name:"merges_file"},{anchor:"transformers.CohereTokenizer.tokenizer_file",description:`<strong>tokenizer_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
<a href="https://github.com/huggingface/tokenizers" rel="nofollow">tokenizers</a> file (generally has a .json extension) that
contains everything needed to load the tokenizer.`,name:"tokenizer_file"},{anchor:"transformers.CohereTokenizer.clean_up_tokenization_spaces",description:`<strong>clean_up_tokenization_spaces</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not to cleanup spaces after decoding, cleanup consists in removing potential artifacts like
extra spaces.`,name:"clean_up_tokenization_spaces"},{anchor:"transformers.CohereTokenizer.unk_token",description:`<strong>unk_token</strong> (<code>str</code> or <code>tokenizers.AddedToken</code>, <em>optional</em>, defaults to <code>&quot;&lt;UNK&gt;&quot;</code>) &#x2014;
The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
token instead.`,name:"unk_token"},{anchor:"transformers.CohereTokenizer.bos_token",description:`<strong>bos_token</strong> (<code>str</code> or <code>tokenizers.AddedToken</code>, <em>optional</em>, defaults to <code>&quot;&lt;BOS_TOKEN&gt;&quot;</code>) &#x2014;
The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.`,name:"bos_token"},{anchor:"transformers.CohereTokenizer.eos_token",description:`<strong>eos_token</strong> (<code>str</code> or <code>tokenizers.AddedToken</code>, <em>optional</em>, defaults to <code>&quot;&lt;|END_OF_TURN_TOKEN|&gt;&quot;</code>) &#x2014;
The end of sequence token.`,name:"eos_token"},{anchor:"transformers.CohereTokenizer.add_bos_token",description:`<strong>add_bos_token</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether or not to add an <code>bos_token</code> at the start of sequences.`,name:"add_bos_token"},{anchor:"transformers.CohereTokenizer.add_eos_token",description:`<strong>add_eos_token</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not to add an <code>eos_token</code> at the end of sequences.`,name:"add_eos_token"},{anchor:"transformers.CohereTokenizer.use_default_system_prompt",description:`<strong>use_default_system_prompt</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not the default system prompt for Cohere tokenizer should be used.`,name:"use_default_system_prompt"},{anchor:"transformers.CohereTokenizer.add_prefix_space",description:`<strong>add_prefix_space</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not the tokenizer should automatically add a prefix space`,name:"add_prefix_space"},{anchor:"transformers.CohereTokenizer.vocab",description:`<strong>vocab</strong> (<code>str</code>, <code>dict</code> or <code>list</code>, <em>optional</em>) &#x2014;
Custom vocabulary dictionary. If not provided, vocabulary is loaded from vocab_file.`,name:"vocab"},{anchor:"transformers.CohereTokenizer.merges",description:`<strong>merges</strong> (<code>str</code> or <code>list[str]</code>, <em>optional</em>) &#x2014;
Custom merges list. If not provided, merges are loaded from <code>merges_file</code>.`,name:"merges"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/tokenization_cohere.py#L45"}}),P=new nt({props:{anchor:"transformers.CohereTokenizer.example",$$slots:{default:[pn]},$$scope:{ctx:J}}}),D=new ho({props:{$$slots:{default:[mn]},$$scope:{ctx:J}}}),Te=new ne({props:{name:"apply_grounded_generation_template",anchor:"transformers.CohereTokenizer.apply_grounded_generation_template",parameters:[{name:"conversation",val:": list"},{name:"documents",val:": list"},{name:"citation_mode",val:": typing.Literal['fast', 'accurate'] = 'accurate'"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.conversation",description:`<strong>conversation</strong> (list[dict[str, str]]) &#x2014; A list of dicts
with &#x201C;role&#x201D; and &#x201C;content&#x201D; keys, representing the chat history so far.`,name:"conversation"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.documents",description:`<strong>documents</strong> (list[dict[str, str]) &#x2014; A list of dicts, representing documents or tool outputs to ground your
generation on. A document is a semistructured dict, with a string to string mapping. Common fields are
<code>url</code>, <code>title</code>, <code>snippet</code> etc but should be descriptive of the key. They will get rendered into the prompt.`,name:"documents"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.citation_mode",description:`<strong>citation_mode</strong> &#x2014; either &#x201C;accurate&#x201D; (prompt the model to generate an answer first, then rewrite it with citation
spans in) or &#x201C;fast&#x201D;, where the prompt instructs the model to generate an answer with citations in directly.
The former has higher quality citations, the latter requires fewer tokens to be generated.`,name:"citation_mode"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.add_generation_prompt",description:`<strong>add_generation_prompt</strong> (bool, <em>optional</em>) &#x2014; Whether to end the prompt with the token(s) that indicate
the start of an assistant message. This is useful when you want to generate a response from the model.
Note that this argument will be passed to the chat template, and so it must be supported in the
template for this argument to have any effect.`,name:"add_generation_prompt"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.tokenize",description:`<strong>tokenize</strong> (<code>bool</code>, defaults to <code>True</code>) &#x2014;
Whether to tokenize the output. If <code>False</code>, the output will be a string.`,name:"tokenize"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.padding",description:`<strong>padding</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to pad sequences to the maximum length. Has no effect if tokenize is <code>False</code>.`,name:"padding"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.truncation",description:`<strong>truncation</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to truncate sequences at the maximum length. Has no effect if tokenize is <code>False</code>.`,name:"truncation"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.max_length",description:`<strong>max_length</strong> (<code>int</code>, <em>optional</em>) &#x2014;
Maximum length (in tokens) to use for padding or truncation. Has no effect if tokenize is <code>False</code>. If
not specified, the tokenizer&#x2019;s <code>max_length</code> attribute will be used as a default.`,name:"max_length"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.return_tensors",description:`<strong>return_tensors</strong> (<code>str</code> or <a href="/docs/transformers/pr_43265/en/internal/file_utils#transformers.TensorType">TensorType</a>, <em>optional</em>) &#x2014;
If set, will return tensors of a particular framework. Has no effect if tokenize is <code>False</code>. Acceptable
values are:
<ul>
<li><code>&apos;pt&apos;</code>: Return PyTorch <code>torch.Tensor</code> objects.</li>
<li><code>&apos;np&apos;</code>: Return NumPy <code>np.ndarray</code> objects.</li>
</ul>`,name:"return_tensors"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to return a dictionary with named outputs. Has no effect if tokenize is <code>False</code>.`,name:"return_dict"},{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.*tokenizer_kwargs",description:"*<strong>*tokenizer_kwargs</strong> &#x2014; Additional kwargs to pass to the tokenizer.",name:"*tokenizer_kwargs"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/tokenization_cohere.py#L297",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A rendered prompt string.
or if tokenize=True:
<code>list[int]</code>: A list of token ids representing the tokenized chat so far, including control tokens. This
output is ready to pass to the model, either directly or via methods like <code>generate()</code>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>str</code></p>
`}}),K=new nt({props:{anchor:"transformers.CohereTokenizer.apply_grounded_generation_template.example",$$slots:{default:[hn]},$$scope:{ctx:J}}}),be=new ne({props:{name:"apply_tool_use_template",anchor:"transformers.CohereTokenizer.apply_tool_use_template",parameters:[{name:"conversation",val:": list"},{name:"tools",val:": list"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CohereTokenizer.apply_tool_use_template.conversation",description:`<strong>conversation</strong> (list[dict[str, str]]) &#x2014; A list of dicts
with &#x201C;role&#x201D; and &#x201C;content&#x201D; keys, representing the chat history so far.`,name:"conversation"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.tools",description:`<strong>tools</strong> (list[Dict]) &#x2014; a list of tools to render into the prompt for the model to choose from.
See an example at the bottom of the docstring.
The format should be:
<ul>
<li>name (str): The name of the tool to be called. Valid names contain only the characters a-z,
A-Z, 0-9, _ and must not begin with a digit.</li>
<li>description (str): The description of what the tool does, the model uses the description to
choose when and how to call the function.</li>
<li>parameter<em>definitions (list[Dict]): The input parameters of the tool. Accepts a dictionary
where the key is the name of the parameter and the value is the parameter spec.
Valid parameter names contain only the characters a-z, A-Z, 0-9, </em> and must not begin with a digit.
Parameter specs are as follows:
<ul>
<li>description (str): The description of the parameter.</li>
<li>type (str): the type of the parameter - most effective for python builtin data types, such as &#x2018;str&#x2019;, &#x2018;bool&#x2019;</li>
<li>required: boolean: Denotes whether the parameter is always present (required) or not. Defaults to not required.</li>
</ul>
</li>
</ul>`,name:"tools"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.add_generation_prompt",description:`<strong>add_generation_prompt</strong> (bool, <em>optional</em>) &#x2014; Whether to end the prompt with the token(s) that indicate
the start of an assistant message. This is useful when you want to generate a response from the model.
Note that this argument will be passed to the chat template, and so it must be supported in the
template for this argument to have any effect.`,name:"add_generation_prompt"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.tokenize",description:`<strong>tokenize</strong> (<code>bool</code>, defaults to <code>True</code>) &#x2014;
Whether to tokenize the output. If <code>False</code>, the output will be a string.`,name:"tokenize"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.padding",description:`<strong>padding</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to pad sequences to the maximum length. Has no effect if tokenize is <code>False</code>.`,name:"padding"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.truncation",description:`<strong>truncation</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to truncate sequences at the maximum length. Has no effect if tokenize is <code>False</code>.`,name:"truncation"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.max_length",description:`<strong>max_length</strong> (<code>int</code>, <em>optional</em>) &#x2014;
Maximum length (in tokens) to use for padding or truncation. Has no effect if tokenize is <code>False</code>. If
not specified, the tokenizer&#x2019;s <code>max_length</code> attribute will be used as a default.`,name:"max_length"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.return_tensors",description:`<strong>return_tensors</strong> (<code>str</code> or <a href="/docs/transformers/pr_43265/en/internal/file_utils#transformers.TensorType">TensorType</a>, <em>optional</em>) &#x2014;
If set, will return tensors of a particular framework. Has no effect if tokenize is <code>False</code>. Acceptable
values are:
<ul>
<li><code>&apos;pt&apos;</code>: Return PyTorch <code>torch.Tensor</code> objects.</li>
<li><code>&apos;np&apos;</code>: Return NumPy <code>np.ndarray</code> objects.</li>
</ul>`,name:"return_tensors"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to return a dictionary with named outputs. Has no effect if tokenize is <code>False</code>.`,name:"return_dict"},{anchor:"transformers.CohereTokenizer.apply_tool_use_template.*tokenizer_kwargs",description:"*<strong>*tokenizer_kwargs</strong> &#x2014; Additional kwargs to pass to the tokenizer.",name:"*tokenizer_kwargs"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/tokenization_cohere.py#L186",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A rendered prompt string.
or if tokenize=True:
<code>list[int]</code>: A list of token ids representing the tokenized chat so far, including control tokens. This
output is ready to pass to the model, either directly or via methods like <code>generate()</code>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>str</code></p>
`}}),O=new nt({props:{anchor:"transformers.CohereTokenizer.apply_tool_use_template.example",$$slots:{default:[un]},$$scope:{ctx:J}}}),we=new et({props:{title:"CohereModel",local:"transformers.CohereModel",headingTag:"h2"}}),Je=new ne({props:{name:"class transformers.CohereModel",anchor:"transformers.CohereModel",parameters:[{name:"config",val:": CohereConfig"}],parametersDescription:[{anchor:"transformers.CohereModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereConfig">CohereConfig</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_43265/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/modeling_cohere.py#L379"}}),Ce=new ne({props:{name:"forward",anchor:"transformers.CohereModel.forward",parameters:[{name:"input_ids",val:": torch.LongTensor | None = None"},{name:"attention_mask",val:": torch.Tensor | None = None"},{name:"position_ids",val:": torch.LongTensor | None = None"},{name:"past_key_values",val:": transformers.cache_utils.Cache | None = None"},{name:"inputs_embeds",val:": torch.FloatTensor | None = None"},{name:"use_cache",val:": bool | None = None"},{name:"**kwargs",val:": typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs]"}],parametersDescription:[{anchor:"transformers.CohereModel.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_43265/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_43265/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_43265/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.CohereModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CohereModel.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.CohereModel.forward.past_key_values",description:`<strong>past_key_values</strong> (<code>~cache_utils.Cache</code>, <em>optional</em>) &#x2014;
Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used to speed up sequential decoding. This typically consists in the <code>past_key_values</code>
returned by the model at a previous stage of decoding, when <code>use_cache=True</code> or <code>config.use_cache=True</code>.</p>
<p>Only <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.Cache">Cache</a> instance is allowed as input, see our <a href="https://huggingface.co/docs/transformers/en/kv_cache" rel="nofollow">kv cache guide</a>.
If no <code>past_key_values</code> are passed, <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.DynamicCache">DynamicCache</a> will be initialized by default.</p>
<p>The model will output the same cache format that is fed as input.</p>
<p>If <code>past_key_values</code> are used, the user is expected to input only unprocessed <code>input_ids</code> (those that don&#x2019;t
have their past key value states given to this model) of shape <code>(batch_size, unprocessed_length)</code> instead of all <code>input_ids</code>
of shape <code>(batch_size, sequence_length)</code>.`,name:"past_key_values"},{anchor:"transformers.CohereModel.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.CohereModel.forward.use_cache",description:`<strong>use_cache</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
If set to <code>True</code>, <code>past_key_values</code> key value states are returned and can be used to speed up decoding (see
<code>past_key_values</code>).`,name:"use_cache"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/modeling_cohere.py#L396",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_43265/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPast"
>BaseModelOutputWithPast</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereConfig"
>CohereConfig</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_43265/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPast"
>BaseModelOutputWithPast</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),ee=new ho({props:{$$slots:{default:[fn]},$$scope:{ctx:J}}}),ke=new et({props:{title:"CohereForCausalLM",local:"transformers.CohereForCausalLM",headingTag:"h2"}}),Ue=new ne({props:{name:"class transformers.CohereForCausalLM",anchor:"transformers.CohereForCausalLM",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.CohereForCausalLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereForCausalLM">CohereForCausalLM</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_43265/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/modeling_cohere.py#L453"}}),ve=new ne({props:{name:"forward",anchor:"transformers.CohereForCausalLM.forward",parameters:[{name:"input_ids",val:": torch.LongTensor | None = None"},{name:"attention_mask",val:": torch.Tensor | None = None"},{name:"position_ids",val:": torch.LongTensor | None = None"},{name:"past_key_values",val:": transformers.cache_utils.Cache | None = None"},{name:"inputs_embeds",val:": torch.FloatTensor | None = None"},{name:"labels",val:": torch.LongTensor | None = None"},{name:"use_cache",val:": bool | None = None"},{name:"logits_to_keep",val:": int | torch.Tensor = 0"},{name:"**kwargs",val:": typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs]"}],parametersDescription:[{anchor:"transformers.CohereForCausalLM.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_43265/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_43265/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_43265/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.CohereForCausalLM.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CohereForCausalLM.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.CohereForCausalLM.forward.past_key_values",description:`<strong>past_key_values</strong> (<code>~cache_utils.Cache</code>, <em>optional</em>) &#x2014;
Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used to speed up sequential decoding. This typically consists in the <code>past_key_values</code>
returned by the model at a previous stage of decoding, when <code>use_cache=True</code> or <code>config.use_cache=True</code>.</p>
<p>Only <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.Cache">Cache</a> instance is allowed as input, see our <a href="https://huggingface.co/docs/transformers/en/kv_cache" rel="nofollow">kv cache guide</a>.
If no <code>past_key_values</code> are passed, <a href="/docs/transformers/pr_43265/en/internal/generation_utils#transformers.DynamicCache">DynamicCache</a> will be initialized by default.</p>
<p>The model will output the same cache format that is fed as input.</p>
<p>If <code>past_key_values</code> are used, the user is expected to input only unprocessed <code>input_ids</code> (those that don&#x2019;t
have their past key value states given to this model) of shape <code>(batch_size, unprocessed_length)</code> instead of all <code>input_ids</code>
of shape <code>(batch_size, sequence_length)</code>.`,name:"past_key_values"},{anchor:"transformers.CohereForCausalLM.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.CohereForCausalLM.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Labels for computing the masked language modeling loss. Indices should either be in <code>[0, ..., config.vocab_size]</code> or -100 (see <code>input_ids</code> docstring). Tokens with indices set to <code>-100</code> are ignored
(masked), the loss is only computed for the tokens with labels in <code>[0, ..., config.vocab_size]</code>.`,name:"labels"},{anchor:"transformers.CohereForCausalLM.forward.use_cache",description:`<strong>use_cache</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
If set to <code>True</code>, <code>past_key_values</code> key value states are returned and can be used to speed up decoding (see
<code>past_key_values</code>).`,name:"use_cache"},{anchor:"transformers.CohereForCausalLM.forward.logits_to_keep",description:`<strong>logits_to_keep</strong> (<code>Union[int, torch.Tensor]</code>, <em>optional</em>, defaults to <code>0</code>) &#x2014;
If an <code>int</code>, compute logits for the last <code>logits_to_keep</code> tokens. If <code>0</code>, calculate logits for all
<code>input_ids</code> (special case). Only last token logits are needed for generation, and calculating them only for that
token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
If a <code>torch.Tensor</code>, must be 1D corresponding to the indices to keep in the sequence length dimension.
This is useful when using packed tensor format (single dimension for batch and sequence length).`,name:"logits_to_keep"}],source:"https://github.com/huggingface/transformers/blob/vr_43265/src/transformers/models/cohere/modeling_cohere.py#L469",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_43265/en/main_classes/output#transformers.modeling_outputs.CausalLMOutputWithPast"
>CausalLMOutputWithPast</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_43265/en/model_doc/cohere#transformers.CohereConfig"
>CohereConfig</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_43265/en/main_classes/output#transformers.modeling_outputs.CausalLMOutputWithPast"
>CausalLMOutputWithPast</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),te=new ho({props:{$$slots:{default:[gn]},$$scope:{ctx:J}}}),oe=new nt({props:{anchor:"transformers.CohereForCausalLM.forward.example",$$slots:{default:[_n]},$$scope:{ctx:J}}}),je=new nn({props:{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/cohere.md"}}),{c(){t=d("meta"),p=s(),o=d("p"),h=s(),w=d("p"),w.innerHTML=m,u=s(),g(U.$$.fragment),se=s(),B=d("div"),B.innerHTML=tt,st=s(),g(ae.$$.fragment),at=s(),re=d("p"),re.innerHTML=fo,rt=s(),le=d("p"),le.innerHTML=go,lt=s(),A=d("blockquote"),A.innerHTML=_o,it=s(),ie=d("p"),ie.innerHTML=yo,dt=s(),g(Q.$$.fragment),ct=s(),de=d("p"),de.innerHTML=Mo,pt=s(),ce=d("p"),ce.innerHTML=To,mt=s(),g(pe.$$.fragment),ht=s(),me=d("p"),me.innerHTML=bo,ut=s(),g(he.$$.fragment),ft=s(),L=d("div"),L.innerHTML=wo,gt=s(),g(ue.$$.fragment),_t=s(),fe=d("ul"),fe.innerHTML=Jo,yt=s(),g(ge.$$.fragment),Mt=s(),G=d("div"),g(_e.$$.fragment),It=s(),Ie=d("p"),Ie.innerHTML=Co,$t=s(),$e=d("p"),$e.innerHTML=ko,zt=s(),g(Y.$$.fragment),Tt=s(),g(ye.$$.fragment),bt=s(),C=d("div"),g(Me.$$.fragment),xt=s(),ze=d("p"),ze.textContent=Uo,Zt=s(),xe=d("p"),xe.textContent=vo,Gt=s(),g(P.$$.fragment),Ft=s(),Ze=d("p"),Ze.innerHTML=jo,Wt=s(),Ge=d("p"),Ge.innerHTML=Io,Nt=s(),g(D.$$.fragment),Rt=s(),Fe=d("p"),Fe.innerHTML=$o,qt=s(),v=d("div"),g(Te.$$.fragment),Bt=s(),We=d("p"),We.textContent=zo,Vt=s(),Ne=d("p"),Ne.textContent=xo,Ht=s(),Re=d("p"),Re.innerHTML=Zo,St=s(),qe=d("p"),qe.innerHTML=Go,Xt=s(),g(K.$$.fragment),Et=s(),j=d("div"),g(be.$$.fragment),At=s(),Be=d("p"),Be.textContent=Fo,Qt=s(),Ve=d("p"),Ve.textContent=Wo,Lt=s(),He=d("p"),He.innerHTML=No,Yt=s(),Se=d("p"),Se.innerHTML=Ro,Pt=s(),g(O.$$.fragment),wt=s(),g(we.$$.fragment),Jt=s(),I=d("div"),g(Je.$$.fragment),Dt=s(),Xe=d("p"),Xe.textContent=qo,Kt=s(),Ee=d("p"),Ee.innerHTML=Bo,Ot=s(),Ae=d("p"),Ae.innerHTML=Vo,eo=s(),R=d("div"),g(Ce.$$.fragment),to=s(),Qe=d("p"),Qe.innerHTML=Ho,oo=s(),g(ee.$$.fragment),no=s(),Le=d("ul"),Le.innerHTML=So,Ct=s(),g(ke.$$.fragment),kt=s(),$=d("div"),g(Ue.$$.fragment),so=s(),Ye=d("p"),Ye.textContent=Xo,ao=s(),Pe=d("p"),Pe.innerHTML=Eo,ro=s(),De=d("p"),De.innerHTML=Ao,lo=s(),Z=d("div"),g(ve.$$.fragment),io=s(),Ke=d("p"),Ke.innerHTML=Qo,co=s(),g(te.$$.fragment),po=s(),Oe=d("ul"),Oe.innerHTML=Lo,mo=s(),g(oe.$$.fragment),Ut=s(),g(je.$$.fragment),vt=s(),ot=d("p"),this.h()},l(e){const n=en("svelte-u9bgzb",document.head);t=c(n,"META",{name:!0,content:!0}),n.forEach(r),p=a(e),o=c(e,"P",{}),X(o).forEach(r),h=a(e),w=c(e,"P",{"data-svelte-h":!0}),f(w)!=="svelte-1bnmcwb"&&(w.innerHTML=m),u=a(e),_(U.$$.fragment,e),se=a(e),B=c(e,"DIV",{style:!0,"data-svelte-h":!0}),f(B)!=="svelte-1d68sxl"&&(B.innerHTML=tt),st=a(e),_(ae.$$.fragment,e),at=a(e),re=c(e,"P",{"data-svelte-h":!0}),f(re)!=="svelte-zmdpgo"&&(re.innerHTML=fo),rt=a(e),le=c(e,"P",{"data-svelte-h":!0}),f(le)!=="svelte-1rlhpz"&&(le.innerHTML=go),lt=a(e),A=c(e,"BLOCKQUOTE",{class:!0,"data-svelte-h":!0}),f(A)!=="svelte-18uxfcw"&&(A.innerHTML=_o),it=a(e),ie=c(e,"P",{"data-svelte-h":!0}),f(ie)!=="svelte-122tlzn"&&(ie.innerHTML=yo),dt=a(e),_(Q.$$.fragment,e),ct=a(e),de=c(e,"P",{"data-svelte-h":!0}),f(de)!=="svelte-nf5ooi"&&(de.innerHTML=Mo),pt=a(e),ce=c(e,"P",{"data-svelte-h":!0}),f(ce)!=="svelte-1ca5nhg"&&(ce.innerHTML=To),mt=a(e),_(pe.$$.fragment,e),ht=a(e),me=c(e,"P",{"data-svelte-h":!0}),f(me)!=="svelte-w3z5ks"&&(me.innerHTML=bo),ut=a(e),_(he.$$.fragment,e),ft=a(e),L=c(e,"DIV",{class:!0,"data-svelte-h":!0}),f(L)!=="svelte-1dgk9vs"&&(L.innerHTML=wo),gt=a(e),_(ue.$$.fragment,e),_t=a(e),fe=c(e,"UL",{"data-svelte-h":!0}),f(fe)!=="svelte-5hpch7"&&(fe.innerHTML=Jo),yt=a(e),_(ge.$$.fragment,e),Mt=a(e),G=c(e,"DIV",{class:!0});var V=X(G);_(_e.$$.fragment,V),It=a(V),Ie=c(V,"P",{"data-svelte-h":!0}),f(Ie)!=="svelte-1xjjsiu"&&(Ie.innerHTML=Co),$t=a(V),$e=c(V,"P",{"data-svelte-h":!0}),f($e)!=="svelte-1e8815j"&&($e.innerHTML=ko),zt=a(V),_(Y.$$.fragment,V),V.forEach(r),Tt=a(e),_(ye.$$.fragment,e),bt=a(e),C=c(e,"DIV",{class:!0});var k=X(C);_(Me.$$.fragment,k),xt=a(k),ze=c(k,"P",{"data-svelte-h":!0}),f(ze)!=="svelte-xb0zhn"&&(ze.textContent=Uo),Zt=a(k),xe=c(k,"P",{"data-svelte-h":!0}),f(xe)!=="svelte-8f5ey4"&&(xe.textContent=vo),Gt=a(k),_(P.$$.fragment,k),Ft=a(k),Ze=c(k,"P",{"data-svelte-h":!0}),f(Ze)!=="svelte-cnb6q1"&&(Ze.innerHTML=jo),Wt=a(k),Ge=c(k,"P",{"data-svelte-h":!0}),f(Ge)!=="svelte-1afeqmz"&&(Ge.innerHTML=Io),Nt=a(k),_(D.$$.fragment,k),Rt=a(k),Fe=c(k,"P",{"data-svelte-h":!0}),f(Fe)!=="svelte-n0kppq"&&(Fe.innerHTML=$o),qt=a(k),v=c(k,"DIV",{class:!0});var z=X(v);_(Te.$$.fragment,z),Bt=a(z),We=c(z,"P",{"data-svelte-h":!0}),f(We)!=="svelte-1qr3i4z"&&(We.textContent=zo),Vt=a(z),Ne=c(z,"P",{"data-svelte-h":!0}),f(Ne)!=="svelte-mcv28b"&&(Ne.textContent=xo),Ht=a(z),Re=c(z,"P",{"data-svelte-h":!0}),f(Re)!=="svelte-lcyshz"&&(Re.innerHTML=Zo),St=a(z),qe=c(z,"P",{"data-svelte-h":!0}),f(qe)!=="svelte-pea71a"&&(qe.innerHTML=Go),Xt=a(z),_(K.$$.fragment,z),z.forEach(r),Et=a(k),j=c(k,"DIV",{class:!0});var x=X(j);_(be.$$.fragment,x),At=a(x),Be=c(x,"P",{"data-svelte-h":!0}),f(Be)!=="svelte-kc13xj"&&(Be.textContent=Fo),Qt=a(x),Ve=c(x,"P",{"data-svelte-h":!0}),f(Ve)!=="svelte-17hueu1"&&(Ve.textContent=Wo),Lt=a(x),He=c(x,"P",{"data-svelte-h":!0}),f(He)!=="svelte-v8vbpm"&&(He.innerHTML=No),Yt=a(x),Se=c(x,"P",{"data-svelte-h":!0}),f(Se)!=="svelte-7a2giu"&&(Se.innerHTML=Ro),Pt=a(x),_(O.$$.fragment,x),x.forEach(r),k.forEach(r),wt=a(e),_(we.$$.fragment,e),Jt=a(e),I=c(e,"DIV",{class:!0});var F=X(I);_(Je.$$.fragment,F),Dt=a(F),Xe=c(F,"P",{"data-svelte-h":!0}),f(Xe)!=="svelte-jpshns"&&(Xe.textContent=qo),Kt=a(F),Ee=c(F,"P",{"data-svelte-h":!0}),f(Ee)!=="svelte-1gb3c10"&&(Ee.innerHTML=Bo),Ot=a(F),Ae=c(F,"P",{"data-svelte-h":!0}),f(Ae)!=="svelte-hswkmf"&&(Ae.innerHTML=Vo),eo=a(F),R=c(F,"DIV",{class:!0});var H=X(R);_(Ce.$$.fragment,H),to=a(H),Qe=c(H,"P",{"data-svelte-h":!0}),f(Qe)!=="svelte-165jkn3"&&(Qe.innerHTML=Ho),oo=a(H),_(ee.$$.fragment,H),no=a(H),Le=c(H,"UL",{"data-svelte-h":!0}),f(Le)!=="svelte-xardb3"&&(Le.innerHTML=So),H.forEach(r),F.forEach(r),Ct=a(e),_(ke.$$.fragment,e),kt=a(e),$=c(e,"DIV",{class:!0});var W=X($);_(Ue.$$.fragment,W),so=a(W),Ye=c(W,"P",{"data-svelte-h":!0}),f(Ye)!=="svelte-1tm3aen"&&(Ye.textContent=Xo),ao=a(W),Pe=c(W,"P",{"data-svelte-h":!0}),f(Pe)!=="svelte-1gb3c10"&&(Pe.innerHTML=Eo),ro=a(W),De=c(W,"P",{"data-svelte-h":!0}),f(De)!=="svelte-hswkmf"&&(De.innerHTML=Ao),lo=a(W),Z=c(W,"DIV",{class:!0});var N=X(Z);_(ve.$$.fragment,N),io=a(N),Ke=c(N,"P",{"data-svelte-h":!0}),f(Ke)!=="svelte-170xwmj"&&(Ke.innerHTML=Qo),co=a(N),_(te.$$.fragment,N),po=a(N),Oe=c(N,"UL",{"data-svelte-h":!0}),f(Oe)!=="svelte-1yb5o5x"&&(Oe.innerHTML=Lo),mo=a(N),_(oe.$$.fragment,N),N.forEach(r),W.forEach(r),Ut=a(e),_(je.$$.fragment,e),vt=a(e),ot=c(e,"P",{}),X(ot).forEach(r),this.h()},h(){q(t,"name","hf:doc:metadata"),q(t,"content",Mn),tn(B,"float","right"),q(A,"class","tip"),q(L,"class","flex justify-center"),q(G,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(v,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(j,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(C,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(R,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(I,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q(Z,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),q($,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(e,n){l(document.head,t),i(e,p,n),i(e,o,n),i(e,h,n),i(e,w,n),i(e,u,n),y(U,e,n),i(e,se,n),i(e,B,n),i(e,st,n),y(ae,e,n),i(e,at,n),i(e,re,n),i(e,rt,n),i(e,le,n),i(e,lt,n),i(e,A,n),i(e,it,n),i(e,ie,n),i(e,dt,n),y(Q,e,n),i(e,ct,n),i(e,de,n),i(e,pt,n),i(e,ce,n),i(e,mt,n),y(pe,e,n),i(e,ht,n),i(e,me,n),i(e,ut,n),y(he,e,n),i(e,ft,n),i(e,L,n),i(e,gt,n),y(ue,e,n),i(e,_t,n),i(e,fe,n),i(e,yt,n),y(ge,e,n),i(e,Mt,n),i(e,G,n),y(_e,G,null),l(G,It),l(G,Ie),l(G,$t),l(G,$e),l(G,zt),y(Y,G,null),i(e,Tt,n),y(ye,e,n),i(e,bt,n),i(e,C,n),y(Me,C,null),l(C,xt),l(C,ze),l(C,Zt),l(C,xe),l(C,Gt),y(P,C,null),l(C,Ft),l(C,Ze),l(C,Wt),l(C,Ge),l(C,Nt),y(D,C,null),l(C,Rt),l(C,Fe),l(C,qt),l(C,v),y(Te,v,null),l(v,Bt),l(v,We),l(v,Vt),l(v,Ne),l(v,Ht),l(v,Re),l(v,St),l(v,qe),l(v,Xt),y(K,v,null),l(C,Et),l(C,j),y(be,j,null),l(j,At),l(j,Be),l(j,Qt),l(j,Ve),l(j,Lt),l(j,He),l(j,Yt),l(j,Se),l(j,Pt),y(O,j,null),i(e,wt,n),y(we,e,n),i(e,Jt,n),i(e,I,n),y(Je,I,null),l(I,Dt),l(I,Xe),l(I,Kt),l(I,Ee),l(I,Ot),l(I,Ae),l(I,eo),l(I,R),y(Ce,R,null),l(R,to),l(R,Qe),l(R,oo),y(ee,R,null),l(R,no),l(R,Le),i(e,Ct,n),y(ke,e,n),i(e,kt,n),i(e,$,n),y(Ue,$,null),l($,so),l($,Ye),l($,ao),l($,Pe),l($,ro),l($,De),l($,lo),l($,Z),y(ve,Z,null),l(Z,io),l(Z,Ke),l(Z,co),y(te,Z,null),l(Z,po),l(Z,Oe),l(Z,mo),y(oe,Z,null),i(e,Ut,n),y(je,e,n),i(e,vt,n),i(e,ot,n),jt=!0},p(e,[n]){const V={};n&2&&(V.$$scope={dirty:n,ctx:e}),Q.$set(V);const k={};n&2&&(k.$$scope={dirty:n,ctx:e}),Y.$set(k);const z={};n&2&&(z.$$scope={dirty:n,ctx:e}),P.$set(z);const x={};n&2&&(x.$$scope={dirty:n,ctx:e}),D.$set(x);const F={};n&2&&(F.$$scope={dirty:n,ctx:e}),K.$set(F);const H={};n&2&&(H.$$scope={dirty:n,ctx:e}),O.$set(H);const W={};n&2&&(W.$$scope={dirty:n,ctx:e}),ee.$set(W);const N={};n&2&&(N.$$scope={dirty:n,ctx:e}),te.$set(N);const Yo={};n&2&&(Yo.$$scope={dirty:n,ctx:e}),oe.$set(Yo)},i(e){jt||(M(U.$$.fragment,e),M(ae.$$.fragment,e),M(Q.$$.fragment,e),M(pe.$$.fragment,e),M(he.$$.fragment,e),M(ue.$$.fragment,e),M(ge.$$.fragment,e),M(_e.$$.fragment,e),M(Y.$$.fragment,e),M(ye.$$.fragment,e),M(Me.$$.fragment,e),M(P.$$.fragment,e),M(D.$$.fragment,e),M(Te.$$.fragment,e),M(K.$$.fragment,e),M(be.$$.fragment,e),M(O.$$.fragment,e),M(we.$$.fragment,e),M(Je.$$.fragment,e),M(Ce.$$.fragment,e),M(ee.$$.fragment,e),M(ke.$$.fragment,e),M(Ue.$$.fragment,e),M(ve.$$.fragment,e),M(te.$$.fragment,e),M(oe.$$.fragment,e),M(je.$$.fragment,e),jt=!0)},o(e){T(U.$$.fragment,e),T(ae.$$.fragment,e),T(Q.$$.fragment,e),T(pe.$$.fragment,e),T(he.$$.fragment,e),T(ue.$$.fragment,e),T(ge.$$.fragment,e),T(_e.$$.fragment,e),T(Y.$$.fragment,e),T(ye.$$.fragment,e),T(Me.$$.fragment,e),T(P.$$.fragment,e),T(D.$$.fragment,e),T(Te.$$.fragment,e),T(K.$$.fragment,e),T(be.$$.fragment,e),T(O.$$.fragment,e),T(we.$$.fragment,e),T(Je.$$.fragment,e),T(Ce.$$.fragment,e),T(ee.$$.fragment,e),T(ke.$$.fragment,e),T(Ue.$$.fragment,e),T(ve.$$.fragment,e),T(te.$$.fragment,e),T(oe.$$.fragment,e),T(je.$$.fragment,e),jt=!1},d(e){e&&(r(p),r(o),r(h),r(w),r(u),r(se),r(B),r(st),r(at),r(re),r(rt),r(le),r(lt),r(A),r(it),r(ie),r(dt),r(ct),r(de),r(pt),r(ce),r(mt),r(ht),r(me),r(ut),r(ft),r(L),r(gt),r(_t),r(fe),r(yt),r(Mt),r(G),r(Tt),r(bt),r(C),r(wt),r(Jt),r(I),r(Ct),r(kt),r($),r(Ut),r(vt),r(ot)),r(t),b(U,e),b(ae,e),b(Q,e),b(pe,e),b(he,e),b(ue,e),b(ge,e),b(_e),b(Y),b(ye,e),b(Me),b(P),b(D),b(Te),b(K),b(be),b(O),b(we,e),b(Je),b(Ce),b(ee),b(ke,e),b(Ue),b(ve),b(te),b(oe),b(je,e)}}}const Mn='{"title":"Cohere","local":"cohere","sections":[{"title":"Notes","local":"notes","sections":[],"depth":2},{"title":"CohereConfig","local":"transformers.CohereConfig","sections":[],"depth":2},{"title":"CohereTokenizer","local":"transformers.CohereTokenizer","sections":[],"depth":2},{"title":"CohereModel","local":"transformers.CohereModel","sections":[],"depth":2},{"title":"CohereForCausalLM","local":"transformers.CohereForCausalLM","sections":[],"depth":2}],"depth":1}';function Tn(J){return Do(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class $n extends Ko{constructor(t){super(),Oo(this,t,Tn,yn,Po,{})}}export{$n as component};