Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74917.66613977983,
-      "required_pages_per_second_upper_estimate": 0.0064936734239072254
+      "required_token_positions_per_second": 74811.30148559465,
+      "required_pages_per_second_upper_estimate": 0.006484454005261205
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9364.708267472479,
-      "required_pages_per_second_upper_estimate": 0.0037357221427606825
+      "required_token_positions_per_second": 9351.412685699332,
+      "required_pages_per_second_upper_estimate": 0.003730418336404712
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18729.416534944958,
-      "required_pages_per_second_upper_estimate": 0.004332720888886643
+      "required_token_positions_per_second": 18702.825371398663,
+      "required_pages_per_second_upper_estimate": 0.004326569491188702
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28094.124802417435,
-      "required_pages_per_second_upper_estimate": 0.0017107848019859383
+      "required_token_positions_per_second": 28054.23805709799,
+      "required_pages_per_second_upper_estimate": 0.0017083559084655638
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28094.124802417435,
-      "required_pages_per_second_upper_estimate": 0.0021627634968050175
+      "required_token_positions_per_second": 28054.23805709799,
+      "required_pages_per_second_upper_estimate": 0.002159692904736749
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28094.124802417435,
-      "required_pages_per_second_upper_estimate": 0.348835005058761
+      "required_token_positions_per_second": 28054.23805709799,
+      "required_pages_per_second_upper_estimate": 0.34833974517424277
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06713546349229205,
+    "producer_batch_seconds": 0.08651237651574206,
     "producer_p95_seconds": 0.22381268100070884,
-    "gpu_update_seconds": 0.9841203523670883,
-    "producer_batches_per_second": 14.89525726019322,
-    "gpu_batches_per_second": 1.0161358797170659,
-    "producer_headroom": 14.658725823499783,
+    "gpu_update_seconds": 0.9855195476608137,
+    "producer_batches_per_second": 11.559039761415372,
+    "gpu_batches_per_second": 1.0146932167642504,
+    "producer_headroom": 11.391659637063437,
     "sustainable": true,
-    "buffer_seconds": 60.031341494392386,
+    "buffer_seconds": 60.116692407309635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 192845.64916735142,
-    "upload_bytes_per_second": 6130107.674552384,
+    "generation_bytes_per_second": 191581.5278163701,
+    "upload_bytes_per_second": 6091143.954698812,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 41.38018709435851,
+    "estimated_full_shards_per_hour": 41.10893607967689,
     "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": 0.9880495930010511,
-      "bytes": 8851756,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 1.0367009409983439,
       "bytes": 10231090,
@@ -562,6 +556,12 @@
       "bytes": 5730133,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9958830260002287,
+      "bytes": 6097937,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 18100, "loss": 0.7005441393703222, "lr": 0.0003, "grad_norm": 0.14222939985207977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2648334775, "tokens_per_second": 149106.81930044683, "data_wait_s": 9.606999810785055e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 18200, "loss": 0.6414412781596184, "lr": 0.0003, "grad_norm": 0.13035340825862457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2662978750, "tokens_per_second": 149118.26991795647, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18300, "loss": 0.7021921593695879, "lr": 0.0003, "grad_norm": 0.1532839326296825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2677681137, "tokens_per_second": 149770.87484752035, "data_wait_s": 1.052200241247192e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 18400, "loss": 0.6478817276656628, "lr": 0.0003, "grad_norm": 0.14570998880915342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2692390728, "tokens_per_second": 149680.40948064916, "data_wait_s": 1.0278999980073422e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff170cac0e44dba735bab27e48382ef1eba9ef35616677e0696104ee210ea204
-size 384334
+oid sha256:b630fa3c610529f0ce924448d7b74b9f667a0236d311b69e196cc1db15925f87
+size 386452