{"parameters":2000000,"checkpointStep":700,"checkpointStage":"refinement","parentStep":2000,"testLoss":0.23577325316146017,"perplexity":1.2658872416921736,"exactMatches":4,"total":40,"outputs":[{"question":"How does Pebble work?","expected":"Pebble is a small programming language.","generated":"Pebble is a small programming language.","exact":true},{"question":"How does let work?","expected":"Use let to declare a mutable variable.","generated":"Todits akenghe usimombentin ro prots.","exact":false},{"question":"How does const work?","expected":"A const binding cannot be reassigned.","generated":"Tons are the units a language model cess.","exact":false},{"question":"How does nil work?","expected":"Nil represents the absence of a value.","generated":"MIT permits reuse with the license notice.","exact":false},{"question":"How does lists work?","expected":"Lists are ordered collections of values.","generated":"Tokens are the units a language model processses.","exact":false},{"question":"How does dictionaries work?","expected":"Dictionaries associate keys with values.","generated":"ints a the lepreductice bin value.","exact":false},{"question":"How does functions work?","expected":"Functions accept arguments and return values.","generated":"Functions accept arguments and return values.","exact":true},{"question":"How does closures work?","expected":"Closures retain their lexical bindings.","generated":"ang predicts to ve areducts.","exact":false},{"question":"How does modules work?","expected":"Modules share explicit exports between files.","generated":"Modus pleshare explices exports between files.","exact":false},{"question":"How does classes work?","expected":"Classes create instances with fields and methods.","generated":"Clss Loss mepres how wred dictions.","exact":false},{"question":"How does loops work?","expected":"Loops repeat a block of statements.","generated":"MIT permits reuse the license notice.","exact":false},{"question":"How does errors work?","expected":"Use try and catch to handle runtime errors.","generated":"Tens re the units langung model processes.","exact":false},{"question":"How does tensors work?","expected":"Tensors store numbers in multidimensional arrays.","generated":"Tokens are the units a language model processs.","exact":false},{"question":"How does parameters work?","expected":"Parameters are weights learned during training.","generated":"Tenth us the units a language model process.","exact":false},{"question":"How does attention work?","expected":"Attention combines information from earlier tokens.","generated":"Tons are the units a language model processes.","exact":false},{"question":"How does training work?","expected":"Training updates weights to reduce prediction loss.","generated":"Tons areding aputed ts the roraictions.","exact":false},{"question":"How does loss work?","expected":"Loss measures how wrong a prediction is.","generated":"Modus pores wred wth the unilices.","exact":false},{"question":"How does gradients work?","expected":"Gradients describe how weights affect the loss.","generated":"Tokens are the units a langunge model process.","exact":false},{"question":"How does AdamW work?","expected":"AdamW updates weights using moments and decay.","generated":"Adits redombers intinto car ther sum.","exact":false},{"question":"How does tokens work?","expected":"Tokens are the units a language model processes.","generated":"Tokens are the units a language model processs.","exact":false},{"question":"How does context work?","expected":"Context is the sequence available to the model.","generated":"Mons porepledens wctiiiinumes prontiles.","exact":false},{"question":"How does softmax work?","expected":"Softmax converts scores into probabilities.","generated":"Aff comautes gradients through tensor operations.","exact":false},{"question":"How does seeds work?","expected":"A seed makes random choices reproducible.","generated":"MIT prmits reuse with the license notice.","exact":false},{"question":"How does JSON work?","expected":"JSON encodes structured data as text.","generated":"sencomas tredints tho a sexprots ba ilinties.","exact":false},{"question":"How does exports work?","expected":"Exports expose declarations to other modules.","generated":"Tokens are the units a language model processes.","exact":false},{"question":"How does MIT work?","expected":"MIT permits reuse with the license notice.","generated":"MIT permits reuse with the license notice.","exact":true},{"question":"How does addition work?","expected":"Addition combines two numbers into their sum.","generated":"Addition combines to to theddlece s.","exact":false},{"question":"How does CPU work?","expected":"A CPU executes general computational instructions.","generated":"A CTextecutes general computintrons iructions.","exact":false},{"question":"How does WASM work?","expected":"WASM runs portable compiled numerical kernels.","generated":"MIT pormits reuse with the license notice.","exact":false},{"question":"How does checkpoints work?","expected":"Checkpoints preserve learned model weights.","generated":"Checkpoints preserve learned model weights.","exact":true},{"question":"How does autodiff work?","expected":"Autodiff computes gradients through tensor operations.","generated":"Autoncombines to numbersintintio nr theior s.","exact":false},{"question":"How does sampling work?","expected":"Sampling chooses a token from model probabilities.","generated":"plints ares the of cctioss from sentoken vabalies.","exact":false},{"question":"How does causal masking work?","expected":"Causal masking prevents access to future tokens.","generated":"Caus masang prevents acctss to future tokens.","exact":false},{"question":"How does embeddings work?","expected":"Embeddings map token IDs to learned vectors.","generated":"IDs to the reducedin ctirns valuections.","exact":false},{"question":"How does normalization work?","expected":"Normalization stabilizes the scale of activations.","generated":"Nonts rmabilizes the scale of activations.","exact":false},{"question":"How does residuals work?","expected":"Residual connections add an input to a layer output.","generated":"Adits redusinghombeddin cays inpumerratings.","exact":false},{"question":"How does perplexity work?","expected":"Perplexity is the exponential of cross entropy.","generated":"is theng momomsbentin pro sctinsele.","exact":false},{"question":"How does validation work?","expected":"Validation measures loss without updating weights.","generated":"Aution pdights axporonts serotweictions.","exact":false},{"question":"How does overfitting work?","expected":"Overfitting memorizes training patterns too closely.","generated":"Atont akents a to uleroucelec.","exact":false},{"question":"How does temperature work?","expected":"Temperature controls the sharpness of sampling.","generated":"Tents are the lenssealg avarng.","exact":false}]}