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": 77125.70549117064,
-      "required_pages_per_second_upper_estimate": 0.005335735262171289
+      "required_token_positions_per_second": 77160.36979438378,
+      "required_pages_per_second_upper_estimate": 0.005338133419099839
     },
     {
       "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": 9640.71318639633,
-      "required_pages_per_second_upper_estimate": 0.004036365290159839
+      "required_token_positions_per_second": 9645.046224297972,
+      "required_pages_per_second_upper_estimate": 0.004038179442645136
     },
     {
       "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": 19281.42637279266,
-      "required_pages_per_second_upper_estimate": 0.004015812628045097
+      "required_token_positions_per_second": 19290.092448595944,
+      "required_pages_per_second_upper_estimate": 0.004017617543095134
     },
     {
       "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": 28922.13955918899,
-      "required_pages_per_second_upper_estimate": 0.0017338381050233096
+      "required_token_positions_per_second": 28935.138672893914,
+      "required_pages_per_second_upper_estimate": 0.0017346173820414225
     },
     {
       "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": 28922.13955918899,
-      "required_pages_per_second_upper_estimate": 0.0022485691356795238
+      "required_token_positions_per_second": 28935.138672893914,
+      "required_pages_per_second_upper_estimate": 0.0022495797595930233
     },
     {
       "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": 28922.13955918899,
-      "required_pages_per_second_upper_estimate": 0.3709821520912891
+      "required_token_positions_per_second": 28935.138672893914,
+      "required_pages_per_second_upper_estimate": 0.371148890764535
     }
   ],
   "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": 0.2789654328821598,
-    "producer_p95_seconds": 0.3613663889991585,
-    "gpu_update_seconds": 0.9559458747309663,
-    "producer_batches_per_second": 3.5846735191109453,
-    "gpu_batches_per_second": 1.0460843301211296,
-    "producer_headroom": 3.426753862851444,
+    "producer_batch_seconds": 0.13281033687530908,
+    "producer_p95_seconds": 0.34913609299837844,
+    "gpu_update_seconds": 0.9555164159589914,
+    "producer_batches_per_second": 7.529534398658025,
+    "gpu_batches_per_second": 1.0465544948239986,
+    "producer_headroom": 7.1945937224456555,
     "sustainable": true,
-    "buffer_seconds": 60.224590108050876,
+    "buffer_seconds": 60.19753420541646,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337194.06884666724,
-    "upload_bytes_per_second": 4500702.288970114,
+    "generation_bytes_per_second": 336383.6285514166,
+    "upload_bytes_per_second": 4580998.3744298415,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.35399769830717,
+    "estimated_full_shards_per_hour": 72.18009607703088,
     "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": 1.2172141400005785,
-      "bytes": 2428421,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.3105249759973958,
       "bytes": 6190634,
@@ -562,6 +556,12 @@
       "bytes": 5147149,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.5611050400038948,
+      "bytes": 11608998,
+      "pages": 27,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -409,3 +409,4 @@
 {"step": 645300, "loss": 0.5486984252929688, "lr": 0.0001, "grad_norm": 0.17951375196020838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94455290557, "tokens_per_second": 151987.93804769317, "data_wait_s": 2.100800338666886e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645400, "loss": 0.5167580842971802, "lr": 0.0001, "grad_norm": 0.16668514075716886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94470019608, "tokens_per_second": 140380.28128611538, "data_wait_s": 7.217000529635698e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 645500, "loss": 0.5505189299583435, "lr": 0.0001, "grad_norm": 0.17340860427545055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94484547368, "tokens_per_second": 150862.3410131202, "data_wait_s": 1.1018004443030804e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 645600, "loss": 0.5486344695091248, "lr": 0.0001, "grad_norm": 0.18047454176415223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94499089731, "tokens_per_second": 151105.4951900034, "data_wait_s": 1.2395998055581003e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da51494136603433656b3ff2616edc4d3e9105564d96ca42e57005a8aa91d3dd
-size 877710
+oid sha256:c7833bf58c881165dfffabf210b953ce59bea1a7c7f378159232057ad801bc68
+size 879828