{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"yi-34b","slug":"yi-34b","name":"Yi-34B","type":"model","url":"https://www.01.ai/","page_url":"https://unfragile.ai/yi-34b","categories":["model-training"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"yi-34b__cap_0","uri":"capability://text.generation.language.bilingual.dense.transformer.inference.with.34b.parameters","name":"bilingual dense transformer inference with 34b parameters","description":"A 34-billion parameter decoder-only transformer model trained on 3 trillion tokens with native support for both English and Chinese language understanding and generation. The model uses standard transformer architecture with optimized attention mechanisms for efficient inference across both languages, leveraging balanced training data to maintain competitive performance in each language without degradation. Implements a unified vocabulary and embedding space that allows seamless code-switching and cross-lingual reasoning within single prompts.","intents":["I need a performant open-source model that handles both English and Chinese without separate model deployments","I want to build multilingual applications without the latency overhead of model switching or ensemble approaches","I need strong Chinese language capability without sacrificing English performance for my production system"],"best_for":["teams building applications serving Chinese and English-speaking users simultaneously","developers deploying open-source models in resource-constrained environments requiring strong bilingual performance","researchers studying cross-lingual transfer and code-switching in large language models"],"limitations":["Performance on languages outside English/Chinese is unknown and likely degraded due to training data composition","No documented performance breakdown between English and Chinese tasks — claims of 'particularly strong for Chinese' are unverified","Bilingual training may introduce interference effects on specialized domains (e.g., technical Chinese terminology vs English technical terms)"],"requires":["GPU with minimum 68GB VRAM for full precision inference (34B × 2 bytes per parameter + KV cache overhead)","Inference framework supporting transformer models (vLLM, Ollama, llama.cpp, or equivalent)","Apache 2.0 license compliance for commercial deployment"],"input_types":["text (English or Chinese)","mixed-language prompts with code-switching"],"output_types":["text (English or Chinese, matching input language or specified target language)"],"categories":["text-generation-language","multilingual-models"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_1","uri":"capability://text.generation.language.general.knowledge.reasoning.with.76.3.mmlu.performance","name":"general knowledge reasoning with 76.3% mmlu performance","description":"Achieves 76.3% accuracy on the Massive Multitask Language Understanding (MMLU) benchmark, indicating strong performance across 57 diverse knowledge domains including STEM, humanities, social sciences, and professional fields. The model demonstrates broad factual knowledge and reasoning capability across these domains through transformer-based pattern matching and learned world knowledge from the 3 trillion token training corpus. Performance is competitive within the 34B parameter class, positioning it as a capable general-purpose reasoning engine for knowledge-intensive tasks.","intents":["I need a model that can answer factual questions across diverse domains with reasonable accuracy for my knowledge base application","I want to benchmark a model's general reasoning capability before integrating it into a production system","I need to handle multi-domain question-answering without building separate specialized models"],"best_for":["developers building general-purpose Q&A systems, chatbots, and knowledge assistants","teams evaluating open-source models for knowledge-intensive applications","researchers comparing model performance across the 34B parameter class"],"limitations":["MMLU score of 76.3% is the only verified benchmark — no breakdown by domain or difficulty level provided","No documentation of performance variance across the 57 MMLU domains; some domains may perform significantly below average","Benchmark was likely computed on a specific inference setup (batch size, temperature, sampling method) that may not match production conditions","No comparison to other 34B models on identical hardware/settings — relative performance unknown"],"requires":["Prompt engineering for knowledge-intensive tasks (few-shot examples improve performance)","Understanding that 76.3% accuracy means ~24% error rate — unsuitable for safety-critical applications without verification","Inference framework supporting standard transformer inference"],"input_types":["text prompts (questions, reasoning tasks, knowledge queries)"],"output_types":["text (answers, explanations, reasoning chains)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_10","uri":"capability://text.generation.language.zero.shot.and.few.shot.task.generalization.through.in.context.learning","name":"zero-shot and few-shot task generalization through in-context learning","description":"Adapts to new tasks through in-context learning by observing examples in the prompt without parameter updates, enabling the model to generalize to unseen tasks by inferring patterns from provided examples. The transformer attention mechanisms learn to recognize task structure from examples and apply learned patterns to generate appropriate outputs for new instances of the same task.","intents":["Perform classification, extraction, or transformation tasks without fine-tuning by providing examples in the prompt","Adapt to domain-specific terminology or formatting conventions through few-shot examples","Rapidly prototype new applications by demonstrating desired behavior through examples rather than training"],"best_for":["Rapid prototyping scenarios where fine-tuning is impractical or unnecessary","Applications requiring task flexibility where different users may specify different tasks","Low-data scenarios where fine-tuning data is unavailable but examples can be provided in prompts"],"limitations":["Few-shot performance is not quantified — no benchmark data on how many examples are needed for effective task learning or how performance compares to fine-tuned models","In-context learning quality degrades with task complexity — simple classification tasks work well, but complex reasoning or multi-step tasks may require more examples than fit in context window","Example selection and ordering significantly impact performance — no guidance on how to construct effective few-shot prompts","In-context learning is fundamentally limited compared to fine-tuning — model cannot update weights to specialize on task, only infer patterns from examples"],"requires":["Clear task examples demonstrating desired input-output behavior","Task examples formatted consistently and placed early in prompt","Context window sufficient for examples plus new task input (4K base limits example count)"],"input_types":["text (task examples with inputs and outputs)","text (new task input to apply learned pattern to)"],"output_types":["text (output following pattern demonstrated in examples)","structured data (if examples demonstrate structured output format)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_2","uri":"capability://text.generation.language.extended.context.window.inference.with.200k.token.support","name":"extended context window inference with 200k token support","description":"Supports an extended context window variant with 200K token capacity (vs. 4K base variant), enabling processing of long-form documents, multi-turn conversations, and large code repositories within a single inference pass. The extended variant likely uses position interpolation, ALiBi, or similar techniques to extend the context window beyond the base training length without retraining. This allows models to maintain coherence and reference accuracy across significantly longer input sequences, critical for document analysis, code understanding, and multi-document reasoning tasks.","intents":["I need to process entire documents or code files without chunking and losing context between sections","I want to maintain conversation history across 50+ turns without losing early context or requiring summarization","I need to analyze relationships across multiple documents or code files in a single inference pass"],"best_for":["developers building document analysis systems, legal/contract review tools, or research paper summarization","teams implementing long-context RAG systems where maintaining full document context improves retrieval quality","engineers working with large codebases who need to understand cross-file dependencies and patterns"],"limitations":["200K context window performance characteristics are completely undocumented — no benchmarks on long-context tasks provided","Extended context likely introduces latency and memory overhead; no throughput or inference speed data available","Position interpolation or similar extension techniques may degrade performance on tasks requiring precise positional reasoning","No documentation of whether 200K variant uses different training or is post-hoc extension of 4K model","Attention complexity scales quadratically with context length — 200K tokens may require specialized inference optimization (sparse attention, paged attention) that may not be available in all frameworks"],"requires":["GPU with significantly higher VRAM than 4K variant (estimated 100GB+ for full precision 200K inference with batch size 1)","Inference framework supporting long-context inference (vLLM with paged attention, or equivalent optimization)","Awareness that 200K tokens ≈ 150K words — requires careful prompt engineering to stay within limits"],"input_types":["text (up to 200K tokens, approximately 150,000 words)"],"output_types":["text (with maintained context coherence across full input length)"],"categories":["text-generation-language","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_3","uri":"capability://code.generation.editing.competitive.coding.task.performance.with.transformer.architecture","name":"competitive coding task performance with transformer architecture","description":"Demonstrates competitive performance on coding tasks (specific benchmarks undocumented) through transformer-based code understanding and generation. The model processes code as text tokens, leveraging the 3 trillion token training corpus which likely includes substantial code data from public repositories. Coding capability emerges from pretraining without documented specialized code fine-tuning, suggesting the base transformer architecture and training data composition are sufficient for code reasoning, completion, and generation tasks.","intents":["I need a model that can complete code snippets and suggest implementations for common programming tasks","I want to use an open-source model for code review, bug detection, or code explanation without external APIs","I need to generate boilerplate code or refactor existing code in multiple programming languages"],"best_for":["developers building IDE plugins or code completion tools using open-source models","teams implementing code analysis and refactoring systems with privacy requirements (on-premise deployment)","researchers studying code generation and understanding in multilingual models"],"limitations":["Coding performance is described only as 'competitive' with no specific benchmarks (HumanEval, MBPP, CodeXGLUE scores unknown)","No documentation of supported programming languages or performance variance across languages","Bilingual training (English/Chinese) may introduce interference on code tasks where English dominates (e.g., variable naming, documentation conventions)","No evidence of specialized code fine-tuning or instruction-following for code tasks — performance likely lower than code-specialized models (CodeLlama, Codex)","Context window of 4K tokens limits ability to process large files or multi-file reasoning"],"requires":["Understanding of code syntax and semantics for effective prompt engineering","Inference framework supporting transformer models","Awareness that model may generate syntactically valid but semantically incorrect code — requires testing/verification"],"input_types":["code snippets (partial or complete)","natural language descriptions of coding tasks","code with comments or docstrings"],"output_types":["code (multiple programming languages)","code explanations and documentation"],"categories":["code-generation-editing","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_4","uri":"capability://text.generation.language.competitive.mathematical.reasoning.with.transformer.based.arithmetic","name":"competitive mathematical reasoning with transformer-based arithmetic","description":"Demonstrates competitive performance on mathematical reasoning tasks (specific benchmarks undocumented) through transformer-based pattern matching and learned mathematical relationships. The model processes mathematical notation and reasoning as text tokens, leveraging training data that includes mathematical problems, proofs, and explanations. Mathematical capability emerges from pretraining without documented specialized math fine-tuning or chain-of-thought training, relying on the transformer's ability to learn mathematical patterns and reasoning from examples in the training corpus.","intents":["I need a model that can solve math problems and explain mathematical reasoning for educational or tutoring applications","I want to use an open-source model for mathematical problem-solving without relying on external APIs","I need to generate mathematical explanations and step-by-step solutions for diverse problem types"],"best_for":["developers building educational tools, tutoring systems, or homework assistance applications","teams implementing mathematical problem-solving systems with privacy requirements","researchers studying mathematical reasoning in large language models"],"limitations":["Mathematical performance is described only as 'competitive' with no specific benchmarks (MATH, GSM8K, SVAMP scores unknown)","No documentation of performance across different mathematical domains (algebra, geometry, calculus, number theory, etc.)","Transformer-based arithmetic is known to struggle with multi-digit calculations and precise numerical reasoning — likely limitation not documented","No evidence of chain-of-thought training or specialized math fine-tuning — performance likely lower than math-specialized models","Bilingual training may introduce interference on mathematical notation and conventions","4K context window limits ability to solve complex multi-step problems requiring extensive working space"],"requires":["Prompt engineering with explicit step-by-step reasoning instructions to improve performance","Verification of numerical answers — model may generate plausible-sounding but incorrect calculations","Inference framework supporting transformer models"],"input_types":["mathematical problems (text or LaTeX notation)","equations and expressions","word problems"],"output_types":["mathematical solutions and explanations","step-by-step reasoning","equations and proofs"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_5","uri":"capability://text.generation.language.apache.2.0.licensed.open.source.model.distribution.and.deployment","name":"apache 2.0 licensed open-source model distribution and deployment","description":"Distributed under Apache 2.0 license, enabling unrestricted commercial use, modification, and redistribution of model weights and architecture. The permissive license allows developers to integrate Yi-34B into proprietary products, fine-tune for specialized domains, and deploy in any environment (cloud, on-premise, edge) without licensing fees or usage restrictions. This open-source distribution model contrasts with closed-source commercial APIs and enables full model ownership and customization for organizations with specific requirements.","intents":["I need to deploy a model in a proprietary product without licensing restrictions or usage fees","I want to fine-tune a model on proprietary data without sharing data with external providers","I need to run a model on-premise or in air-gapped environments for compliance or security reasons"],"best_for":["enterprises with strict data privacy requirements or compliance obligations (HIPAA, GDPR, etc.)","teams building proprietary products who need model ownership without licensing complexity","organizations deploying models in regulated industries (finance, healthcare) where external API dependencies are problematic","developers who want to fine-tune models on proprietary data without cloud provider lock-in"],"limitations":["Apache 2.0 license requires attribution — must include license and copyright notice in distributions","No commercial support or SLA from 01.AI — organizations must manage deployment, optimization, and troubleshooting independently","Open-source distribution means no guaranteed security updates or vulnerability patches — organizations responsible for monitoring and updating","No hosted inference endpoint from 01.AI — requires self-managed infrastructure (GPU servers, inference frameworks, scaling)","Community-driven support only — no guaranteed response times for issues or feature requests"],"requires":["Understanding of Apache 2.0 license terms and attribution requirements","Infrastructure for model deployment (GPU servers, inference framework, monitoring)","Technical expertise for model optimization, fine-tuning, and troubleshooting","Compliance review to ensure open-source license is acceptable in organization's context"],"input_types":["model weights (in supported formats: GGUF, SafeTensors, etc.)","training data for fine-tuning"],"output_types":["fine-tuned model weights","inference outputs (text)"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_6","uri":"capability://text.generation.language.foundation.model.for.downstream.fine.tuning.and.specialized.adaptation","name":"foundation model for downstream fine-tuning and specialized adaptation","description":"Serves as a foundation model for creating specialized variants through instruction tuning, domain-specific fine-tuning, and alignment training. The 34B base model provides a strong starting point for organizations to adapt to specific use cases (customer service, medical diagnosis, legal analysis, etc.) without training from scratch. This capability is evidenced by Yi-34B's role as the foundation for Yi-1.5 and subsequent models from 01.AI, demonstrating the model's suitability for downstream adaptation and specialization.","intents":["I want to fine-tune a model on my domain-specific data without training from scratch","I need to create specialized variants of a model for different use cases (customer service, medical, legal)","I want to align a model to specific values, tones, or behavioral guidelines through instruction tuning"],"best_for":["organizations with domain-specific data who want to create specialized models without full pretraining","teams building multiple model variants for different customer segments or use cases","researchers studying model adaptation, transfer learning, and instruction tuning techniques"],"limitations":["Fine-tuning methodology and best practices are not documented — organizations must develop their own approaches","No guidance on data requirements, training hyperparameters, or convergence criteria for fine-tuning","Fine-tuning on small datasets may lead to catastrophic forgetting of base model capabilities","No documented instruction tuning data or RLHF methodology — organizations must create their own alignment data","Bilingual base model may introduce interference when fine-tuning for single-language specialization"],"requires":["Domain-specific training data (quantity and quality depend on specialization goals)","Fine-tuning infrastructure (GPU cluster, distributed training framework like DeepSpeed or FSDP)","Expertise in fine-tuning methodology, hyperparameter selection, and evaluation","Evaluation framework to measure specialization success and avoid capability degradation"],"input_types":["base model weights","domain-specific training data (text, instruction-response pairs, or preference data for RLHF)"],"output_types":["fine-tuned model weights","specialized model variants"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_7","uri":"capability://text.generation.language.multilingual.code.switching.and.cross.lingual.reasoning","name":"multilingual code-switching and cross-lingual reasoning","description":"Supports seamless code-switching between English and Chinese within single prompts and responses, enabling cross-lingual reasoning and mixed-language outputs. The unified bilingual architecture processes both languages through a shared vocabulary and embedding space, allowing the model to understand relationships between English and Chinese concepts, translate between languages implicitly, and generate responses that mix both languages naturally. This capability is particularly valuable for applications serving bilingual users or requiring cross-lingual understanding.","intents":["I need a model that can understand mixed English-Chinese prompts and respond appropriately","I want to build applications for bilingual users who naturally code-switch between languages","I need to perform cross-lingual reasoning, such as understanding English technical documentation and explaining it in Chinese"],"best_for":["developers building applications for Chinese and English-speaking users who code-switch naturally","teams implementing cross-lingual search, translation, or understanding systems","organizations serving bilingual markets (China, Singapore, Taiwan, diaspora communities)"],"limitations":["Code-switching performance is not documented — no benchmarks on mixed-language understanding or generation","No documentation of how well the model handles language mixing at different granularities (word-level, phrase-level, sentence-level)","Bilingual training may introduce interference where the model confuses similar concepts across languages or generates code-switched output when single-language output is preferred","No guidance on prompt engineering for code-switching tasks — users must experiment to find effective approaches","Performance on other language pairs (English-Spanish, English-French, etc.) is unknown and likely degraded"],"requires":["Understanding of both English and Chinese to evaluate code-switching quality","Prompt engineering to guide the model toward desired language mixing patterns","Inference framework supporting transformer models"],"input_types":["mixed English-Chinese prompts","code-switched text with language mixing at various granularities"],"output_types":["code-switched responses","mixed-language explanations and reasoning"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_8","uri":"capability://text.generation.language.instruction.following.and.task.specific.prompt.adaptation","name":"instruction-following and task-specific prompt adaptation","description":"Responds to natural language instructions and task specifications through learned instruction-following patterns in training data, enabling users to specify desired behavior through prompts without explicit fine-tuning. The model interprets instructions like 'summarize this text', 'translate to Chinese', or 'explain this code' and adapts its output format and content accordingly through attention mechanisms trained on instruction-response pairs.","intents":["Build conversational AI systems where users specify tasks through natural language instructions","Create prompt-based automation workflows for content generation, summarization, and transformation","Enable non-technical users to interact with AI through natural language task descriptions"],"best_for":["Conversational AI and chatbot applications requiring flexible task handling","Content creation and marketing automation tools with diverse task requirements","Internal tools and productivity applications where users specify tasks through prompts"],"limitations":["Instruction-following quality is not quantified — no benchmark data on how well Yi-34B follows complex, multi-step instructions compared to instruction-tuned models","Instruction-following methodology (whether from base training or explicit instruction tuning) is undocumented — unclear if model received dedicated instruction-tuning phase","Prompt sensitivity is unknown — unclear how robust instruction-following is to prompt variations, ambiguity, or adversarial inputs","No guidance on prompt engineering best practices for Yi-34B — users must discover effective prompting strategies through trial and error"],"requires":["Clear, well-formed natural language instructions in English or Chinese","Understanding of model capabilities and limitations for effective prompt design","No special formatting or markup required (though structured prompts may improve results)"],"input_types":["text (natural language instructions)","text (task specifications)","text (content to transform or analyze)"],"output_types":["text (instruction-following responses)","text (task-specific output formats)","structured data (if instruction specifies structured output)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__cap_9","uri":"capability://text.generation.language.multi.turn.conversation.context.management.and.coherence.maintenance","name":"multi-turn conversation context management and coherence maintenance","description":"Maintains conversation state across multiple turns through transformer attention mechanisms that reference previous messages in the conversation history, enabling coherent multi-turn dialogues where the model understands context, pronouns, and references to earlier statements. The model uses positional embeddings and attention patterns to weight recent messages more heavily while retaining access to earlier conversation context.","intents":["Build conversational chatbots that maintain coherent dialogue across 10+ turns without losing context","Create interactive tutoring systems where students can ask follow-up questions and receive contextually appropriate responses","Develop customer support agents that understand conversation history and provide consistent, coherent assistance"],"best_for":["Conversational AI applications requiring natural multi-turn dialogue","Interactive systems where users expect the AI to remember earlier statements and questions","Customer support and helpdesk automation requiring conversation continuity"],"limitations":["Context window limitations (4K base, 200K extended) constrain conversation length — after ~50-100 turns, early conversation context is lost even with 4K window","No explicit conversation memory or summarization — model cannot selectively compress old context to preserve important information while freeing tokens","Coherence degradation over long conversations is not quantified — unclear at what conversation length quality noticeably declines","No built-in conversation state management — applications must manually maintain conversation history and pass it to each inference call"],"requires":["Conversation history formatted as message list (typically alternating user/assistant messages)","Context window sufficient for conversation length (4K base supports ~50-100 turns depending on message length)","Application-level conversation state management (no built-in persistence)"],"input_types":["text (current user message)","text (conversation history from previous turns)"],"output_types":["text (contextually appropriate response)","text (response referencing earlier conversation)"],"categories":["text-generation-language","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"yi-34b__headline","uri":"capability://text.generation.language.bilingual.language.model.for.english.and.chinese","name":"bilingual language model for english and chinese","description":"Yi-34B is a powerful bilingual language model designed for high-performance tasks in both English and Chinese, making it ideal for developers seeking robust language processing capabilities.","intents":["best bilingual language model","language model for Chinese tasks","top AI model for coding and math","high-performance English language model","34 billion parameter language model"],"best_for":["bilingual applications","coding tasks","math problem-solving"],"limitations":[],"requires":[],"input_types":[],"output_types":[],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":57,"verified":false,"data_access_risk":"high","permissions":["GPU with minimum 68GB VRAM for full precision inference (34B × 2 bytes per parameter + KV cache overhead)","Inference framework supporting transformer models (vLLM, Ollama, llama.cpp, or equivalent)","Apache 2.0 license compliance for commercial deployment","Prompt engineering for knowledge-intensive tasks (few-shot examples improve performance)","Understanding that 76.3% accuracy means ~24% error rate — unsuitable for safety-critical applications without verification","Inference framework supporting standard transformer inference","Clear task examples demonstrating desired input-output behavior","Task examples formatted consistently and placed early in prompt","Context window sufficient for examples plus new task input (4K base limits example count)","GPU with significantly higher VRAM than 4K variant (estimated 100GB+ for full precision 200K inference with batch size 1)"],"failure_modes":["Performance on languages outside English/Chinese is unknown and likely degraded due to training data composition","No documented performance breakdown between English and Chinese tasks — claims of 'particularly strong for Chinese' are unverified","Bilingual training may introduce interference effects on specialized domains (e.g., technical Chinese terminology vs English technical terms)","MMLU score of 76.3% is the only verified benchmark — no breakdown by domain or difficulty level provided","No documentation of performance variance across the 57 MMLU domains; some domains may perform significantly below average","Benchmark was likely computed on a specific inference setup (batch size, temperature, sampling method) that may not match production conditions","No comparison to other 34B models on identical hardware/settings — relative performance unknown","Few-shot performance is not quantified — no benchmark data on how many examples are needed for effective task learning or how performance compares to fine-tuned models","In-context learning quality degrades with task complexity — simple classification tasks work well, but complex reasoning or multi-step tasks may require more examples than fit in context window","Example selection and ordering significantly impact performance — no guidance on how to construct effective few-shot prompts","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":0.9,"ecosystem":0.3,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.35,"quality":0.2,"ecosystem":0.1,"match_graph":0.3,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:34.804Z","last_scraped_at":null,"last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=yi-34b","compare_url":"https://unfragile.ai/compare?artifact=yi-34b"}},"signature":"l18krnaavhYEL/H88H67cFYNU1z4csmWk1d0ecPxU+sDwlJP6/xpY3GailFxIcmBnVnrT9hM7F9ypIoSHoyQDg==","signedAt":"2026-06-22T05:25:07.862Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/yi-34b","artifact":"https://unfragile.ai/yi-34b","verify":"https://unfragile.ai/api/v1/verify?slug=yi-34b","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}