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": 77363.93460145993,
-      "required_pages_per_second_upper_estimate": 0.005352216504788795
+      "required_token_positions_per_second": 77214.3061863955,
+      "required_pages_per_second_upper_estimate": 0.005341864863848882
     },
     {
       "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": 9670.491825182491,
-      "required_pages_per_second_upper_estimate": 0.004048832984370914
+      "required_token_positions_per_second": 9651.788273299437,
+      "required_pages_per_second_upper_estimate": 0.004041002197772057
     },
     {
       "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": 19340.983650364982,
-      "required_pages_per_second_upper_estimate": 0.004028216838332383
+      "required_token_positions_per_second": 19303.576546598873,
+      "required_pages_per_second_upper_estimate": 0.004020425925109567
     },
     {
       "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": 29011.475475547475,
-      "required_pages_per_second_upper_estimate": 0.001739193656800955
+      "required_token_positions_per_second": 28955.36481989831,
+      "required_pages_per_second_upper_estimate": 0.0017358299086707978
     },
     {
       "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": 29011.475475547475,
-      "required_pages_per_second_upper_estimate": 0.002255514610229228
+      "required_token_positions_per_second": 28955.36481989831,
+      "required_pages_per_second_upper_estimate": 0.0022511522535571963
     },
     {
       "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": 29011.475475547475,
-      "required_pages_per_second_upper_estimate": 0.3721280573049021
+      "required_token_positions_per_second": 28955.36481989831,
+      "required_pages_per_second_upper_estimate": 0.37140833006116275
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.4772430246016768,
-    "producer_p95_seconds": 0.2899094519998471,
-    "gpu_update_seconds": 0.953002201604786,
-    "producer_batches_per_second": 2.095368498753091,
-    "gpu_batches_per_second": 1.0493155192255308,
-    "producer_headroom": 1.996890792485011,
+    "producer_batch_seconds": 0.09184272381233427,
+    "producer_p95_seconds": 0.26047595300042303,
+    "gpu_update_seconds": 0.9548489605283826,
+    "producer_batches_per_second": 10.888178817990394,
+    "gpu_batches_per_second": 1.0472860539604423,
+    "producer_headroom": 10.396566226405282,
     "sustainable": true,
-    "buffer_seconds": 60.039138701101514,
+    "buffer_seconds": 60.15548451328811,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302911.2564810586,
-    "upload_bytes_per_second": 3861881.8803506247,
+    "generation_bytes_per_second": 303164.14278089185,
+    "upload_bytes_per_second": 3899995.085709593,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.99770422767466,
+    "estimated_full_shards_per_hour": 65.05196774072711,
     "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": 0.8537721579996287,
-      "bytes": 1128490,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.7903945020007086,
       "bytes": 11075879,
@@ -562,6 +556,12 @@
       "bytes": 6797860,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0245990330004133,
+      "bytes": 6034884,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -287,3 +287,4 @@
 {"step": 553700, "loss": 0.5906459093093872, "lr": 0.0003, "grad_norm": 0.15921287635728937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81044091188, "tokens_per_second": 152003.92064805134, "data_wait_s": 9.809999028220773e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 553800, "loss": 0.5602348446846008, "lr": 0.0003, "grad_norm": 0.1597034305431746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81058604879, "tokens_per_second": 134630.7555000232, "data_wait_s": 1.3774002582067624e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553900, "loss": 0.5510703325271606, "lr": 0.0003, "grad_norm": 0.1767396826164066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81073063133, "tokens_per_second": 150096.04300473072, "data_wait_s": 9.92700006463565e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 554000, "loss": 0.5799801349639893, "lr": 0.0003, "grad_norm": 0.14710493437358754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81087696919, "tokens_per_second": 143095.84356363709, "data_wait_s": 1.1952997738262638e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edec37675a78fcbf093b33c33d588749983c556511210c9e0b0040f304370caa
-size 616182
+oid sha256:b88bcac4a8d5b2c05b206eb1ad3e6602dbd9188180eb807d842a8fae34816da1
+size 618300