Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77368.5801133531,
-      "required_pages_per_second_upper_estimate": 0.005352537892080638
+      "required_token_positions_per_second": 76779.29520323357,
+      "required_pages_per_second_upper_estimate": 0.005311769794669196
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9671.072514169138,
-      "required_pages_per_second_upper_estimate": 0.004049076106723459
+      "required_token_positions_per_second": 9597.411900404197,
+      "required_pages_per_second_upper_estimate": 0.004018235945948609
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19342.145028338276,
-      "required_pages_per_second_upper_estimate": 0.004028458722736623
+      "required_token_positions_per_second": 19194.823800808394,
+      "required_pages_per_second_upper_estimate": 0.003997775596164182
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29013.217542507413,
-      "required_pages_per_second_upper_estimate": 0.0017392980910552209
+      "required_token_positions_per_second": 28792.235701212594,
+      "required_pages_per_second_upper_estimate": 0.001726050567089331
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29013.217542507413,
-      "required_pages_per_second_upper_estimate": 0.0022556500482728203
+      "required_token_positions_per_second": 28792.235701212594,
+      "required_pages_per_second_upper_estimate": 0.002238469682108544
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29013.217542507413,
-      "required_pages_per_second_upper_estimate": 0.37215040266937843
+      "required_token_positions_per_second": 28792.235701212594,
+      "required_pages_per_second_upper_estimate": 0.36931588488106354
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13940570900803095,
-    "producer_p95_seconds": 0.31362525500298943,
-    "gpu_update_seconds": 0.9529449796284323,
-    "producer_batches_per_second": 7.173307371094762,
-    "gpu_batches_per_second": 1.0493785280131442,
-    "producer_headroom": 6.835767246616382,
+    "producer_batch_seconds": 0.31539869791396313,
+    "producer_p95_seconds": 0.309170925997023,
+    "gpu_update_seconds": 0.9602588797519325,
+    "producer_batches_per_second": 3.1705901343726777,
+    "gpu_batches_per_second": 1.0413858398876081,
+    "producer_headroom": 3.044587330585237,
     "sustainable": true,
-    "buffer_seconds": 60.03553371659123,
+    "buffer_seconds": 60.496309424371745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352914.8896312424,
-    "upload_bytes_per_second": 4998427.53890997,
+    "generation_bytes_per_second": 339432.8541656326,
+    "upload_bytes_per_second": 4993930.288443111,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.727319876699,
+    "estimated_full_shards_per_hour": 72.83438891150222,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 2.086732669002231,
-      "bytes": 12324271,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.4604756749977241,
       "bytes": 1030943,
@@ -562,6 +556,12 @@
       "bytes": 12000675,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1082588089993806,
+      "bytes": 7050197,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 921600, "loss": 0.46359577775001526, "lr": 3e-05, "grad_norm": 0.32375158862476977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134895709170, "tokens_per_second": 152926.93779968837, "data_wait_s": 1.3755998224951327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921700, "loss": 0.5692005753517151, "lr": 3e-05, "grad_norm": 0.3760321270501417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134910390259, "tokens_per_second": 153010.57421164497, "data_wait_s": 9.431001672055572e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 921800, "loss": 0.5565356612205505, "lr": 3e-05, "grad_norm": 0.3252973719380204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134925109047, "tokens_per_second": 153608.86688052217, "data_wait_s": 1.5084995538927615e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 921900, "loss": 0.525381863117218, "lr": 3e-05, "grad_norm": 0.3327822112093991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134939748858, "tokens_per_second": 152573.92176026368, "data_wait_s": 1.3678996765520424e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:020d2d484aa8c5d8d31a646472e2d3718504095f278e9c80c887ad3bbd21a9fd
-size 785070
+oid sha256:d26ac56165acf076e8efe8aeefe07c9d15044598ae21682c61339b3675097ee4
+size 787188