Charla: La ciencia de la lectura
Aula 1307, pabellón 0 + infinito, viernes, 29 de mayo, 15:00 GMT-3
Charla: La Ciencia de la Lectura. ¿Es posible una revolución en la alfabetización argentina?
Viene Andrés Rieznik, científico y divulgador, a charlar sobre cómo usar Ciencia de Datos para transformar las politicas publicas de alfabetización.
Te esperamos a las 15hs en el aula 1307 del 0+Infinito
https://cartelera.inexactas.ar/event/charla-la-ciencia-de-la-lectura
Reunión docentes y estudiantes DC
espacio 0,5, Ciudad Universitaria, Buenos Aires, martes, 2 de junio, 18:00 GMT-3
Hola a todxs,
Desde la Asamblea Docente del DC, la ComCom y la CODA lxs invitamos a participar de la Asamblea de Docentes y Estudiantes del DC. La idea es construir un espacio en el que podamos hablar horizontalmente entre docentes y estudiantes con el fin de debatir ideas y
medidas que podemos tomar en lo que queda del cuatrimestre y el siguiente.
Nos vemos el martes 2/6 a las 18hs en el espacio entre el pabellón 0+inf y el pabellón 1, en la planta baja.
Asamblea Docente del DC
ComCom
CODA
https://cartelera.inexactas.ar/event/reunion-docentes-y-estudiantes-dc
Transformer training systems are built around dense linear algebra, yet a nontrivial fraction of end-to-end time is spent on surrounding memory-bound operators. Normalization, activations, residual updates, reductions, and related computations repeatedly move large intermediate tensors through global memory while performing little arithmetic, making data movement an increasingly important bottleneck in otherwise highly optimized training stacks. We introduce CODA, a GPU kernel abstraction that expresses these computations as GEMM-plus-epilogue programs. CODA is based on the observation that many Transformer operators exposed as separate framework kernels can be algebraically reparameterized to execute while a GEMM output tile remains on chip, before it is written to memory. The abstraction fixes the GEMM mainloop and exposes a small set of composable epilogue primitives for scaling, reductions, pairwise transformations, and accumulation. This constrained interface preserves the performance structure of expert-written GEMMs while remaining expressive enough to cover nearly all non-attention computation in the forward and backward pass of a standard Transformer block. Across representative Transformer workloads, both human- and LLM-authored CODA kernels achieve high performance, suggesting that GEMM-plus-epilogue programming offers a practical path toward combining framework-level productivity with hardware-level efficiency.
CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs
https://arxiv.org/abs/2605.19269
#HackerNews #CODA #Transformer #GEMM-Epilogue #AI #Research #MachineLearning
Transformer training systems are built around dense linear algebra, yet a nontrivial fraction of end-to-end time is spent on surrounding memory-bound operators. Normalization, activations, residual updates, reductions, and related computations repeatedly move large intermediate tensors through global memory while performing little arithmetic, making data movement an increasingly important bottleneck in otherwise highly optimized training stacks. We introduce CODA, a GPU kernel abstraction that expresses these computations as GEMM-plus-epilogue programs. CODA is based on the observation that many Transformer operators exposed as separate framework kernels can be algebraically reparameterized to execute while a GEMM output tile remains on chip, before it is written to memory. The abstraction fixes the GEMM mainloop and exposes a small set of composable epilogue primitives for scaling, reductions, pairwise transformations, and accumulation. This constrained interface preserves the performance structure of expert-written GEMMs while remaining expressive enough to cover nearly all non-attention computation in the forward and backward pass of a standard Transformer block. Across representative Transformer workloads, both human- and LLM-authored CODA kernels achieve high performance, suggesting that GEMM-plus-epilogue programming offers a practical path toward combining framework-level productivity with hardware-level efficiency.
CODA confirma su regreso a Chile y se suma a la agenda de conciertos japoneses del año | vía #HyperMusicaCL
De Turing a la Super-Inteligencia
Aula 1402, Pabellón 0+∞, martes, 19 de mayo, 15:00 GMT-3
https://cartelera.inexactas.ar/event/de-turing-a-la-super-inteligencia
Emilio Insolera Wants More Deaf Actors Heard, Not Just Seen, in Hollywood Movies: “What Stories We Have to Share”
#MovieNews #Movies #Coda #International #Netflix
The 'Feel My Voice' actor plays a father to a shy teen with a gift for singing in the Netflix drama, a remake of the French film from which 'CODA' was adapted and which features authentic deaf characters.
Vive le Box Office: French Film Elite Launches English-Language Label Emotion Pictures
#Business #BusinessNews #Cannes #Coda #France #International #Pathe #Vendôme
LaRampa.it
Eventos Exactas
N-gated Hacker News
Hacker News
Noticias Rock CL
The Hollywood Reporter
Nika Mavrody
