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": 75998.78075601716,
-      "required_pages_per_second_upper_estimate": 0.005257771994167734
+      "required_token_positions_per_second": 75517.6239183727,
+      "required_pages_per_second_upper_estimate": 0.005224484447701812
     },
     {
       "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": 9499.847594502146,
-      "required_pages_per_second_upper_estimate": 0.003977387808441805
+      "required_token_positions_per_second": 9439.702989796588,
+      "required_pages_per_second_upper_estimate": 0.0039522065184137535
     },
     {
       "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": 18999.69518900429,
-      "required_pages_per_second_upper_estimate": 0.0039571354522129265
+      "required_token_positions_per_second": 18879.405979593175,
+      "required_pages_per_second_upper_estimate": 0.0039320823821323665
     },
     {
       "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": 28499.542783506437,
-      "required_pages_per_second_upper_estimate": 0.001708504073589055
+      "required_token_positions_per_second": 28319.108969389763,
+      "required_pages_per_second_upper_estimate": 0.0016976873419392421
     },
     {
       "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": 28499.542783506437,
-      "required_pages_per_second_upper_estimate": 0.0022157140951769767
+      "required_token_positions_per_second": 28319.108969389763,
+      "required_pages_per_second_upper_estimate": 0.0022016861597738786
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28499.542783506437,
-      "required_pages_per_second_upper_estimate": 0.36556153440189887
+      "required_token_positions_per_second": 28319.108969389763,
+      "required_pages_per_second_upper_estimate": 0.36324712316914565
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1307396596874355,
-    "producer_p95_seconds": 0.3007321229997615,
-    "gpu_update_seconds": 0.9701208264997412,
-    "producer_batches_per_second": 7.648788457846225,
-    "gpu_batches_per_second": 1.0307994351673335,
-    "producer_headroom": 7.42024898044746,
+    "producer_batch_seconds": 0.15657385643768862,
+    "producer_p95_seconds": 0.30418533899864997,
+    "gpu_update_seconds": 0.97630190377405,
+    "producer_batches_per_second": 6.386762277890038,
+    "gpu_batches_per_second": 1.0242733278859144,
+    "producer_headroom": 6.235408170856332,
     "sustainable": true,
-    "buffer_seconds": 60.14749124298395,
+    "buffer_seconds": 60.5307180339911,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407662.69076270325,
-    "upload_bytes_per_second": 5787383.286457806,
+    "generation_bytes_per_second": 431498.2319016139,
+    "upload_bytes_per_second": 5735935.499491359,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.47492353592703,
+    "estimated_full_shards_per_hour": 92.58947580133736,
     "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": 0.972795835998113,
-      "bytes": 8254778,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3151332330016885,
       "bytes": 10337557,
@@ -562,6 +556,12 @@
       "bytes": 7187251,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.2223667069993098,
+      "bytes": 5868379,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 586400, "loss": 0.5779256820678711, "lr": 0.0003, "grad_norm": 0.16024750350676212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85831748706, "tokens_per_second": 150906.63926133554, "data_wait_s": 1.4175999240251258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586500, "loss": 0.5583824515342712, "lr": 0.0003, "grad_norm": 0.17990932988075062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85846333053, "tokens_per_second": 151401.49081931167, "data_wait_s": 9.75199873209931e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 586600, "loss": 0.5461352467536926, "lr": 0.0003, "grad_norm": 0.16201072949906475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85860814427, "tokens_per_second": 149998.77965041617, "data_wait_s": 2.0488001609919593e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 586700, "loss": 0.5536483526229858, "lr": 0.0003, "grad_norm": 0.14367783967544867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85875449788, "tokens_per_second": 152037.35700774894, "data_wait_s": 8.17299951449968e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8aec0dd4e164f98689fdc9e454c5c27a3491deba4671ad69454f53c996e735b
-size 629673
+oid sha256:b19ac0c562d48a8d5953188cb7ba1f5d7a702bc61452638ab080209812e15824
+size 631791