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": 76958.7492344676,
-      "required_pages_per_second_upper_estimate": 0.005324184840940686
+      "required_token_positions_per_second": 76751.58763510919,
+      "required_pages_per_second_upper_estimate": 0.005309852920815935
     },
     {
       "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": 9619.84365430845,
-      "required_pages_per_second_upper_estimate": 0.004027627652880031
+      "required_token_positions_per_second": 9593.948454388648,
+      "required_pages_per_second_upper_estimate": 0.004016785873426875
     },
     {
       "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": 19239.6873086169,
-      "required_pages_per_second_upper_estimate": 0.00400711948171043
+      "required_token_positions_per_second": 19187.896908777297,
+      "required_pages_per_second_upper_estimate": 0.003996332907228531
     },
     {
       "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": 28859.53096292535,
-      "required_pages_per_second_upper_estimate": 0.0017300848152750957
+      "required_token_positions_per_second": 28781.845363165947,
+      "required_pages_per_second_upper_estimate": 0.001725427682188562
     },
     {
       "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": 28859.53096292535,
-      "required_pages_per_second_upper_estimate": 0.002243701592706137
+      "required_token_positions_per_second": 28781.845363165947,
+      "required_pages_per_second_upper_estimate": 0.0022376618790276846
     },
     {
       "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": 28859.53096292535,
-      "required_pages_per_second_upper_estimate": 0.3701790762422923
+      "required_token_positions_per_second": 28781.845363165947,
+      "required_pages_per_second_upper_estimate": 0.36918260878087694
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3333538296249543,
-    "producer_p95_seconds": 0.2098371939991921,
-    "gpu_update_seconds": 0.9580197278853301,
-    "producer_batches_per_second": 2.999815544717359,
-    "gpu_batches_per_second": 1.0438198409622883,
-    "producer_headroom": 2.8738824718563074,
+    "producer_batch_seconds": 0.05365570578901213,
+    "producer_p95_seconds": 0.1881421459984267,
+    "gpu_update_seconds": 0.9606055362726323,
+    "producer_batches_per_second": 18.637346863579694,
+    "gpu_batches_per_second": 1.0410100319432127,
+    "producer_headroom": 17.903138578588035,
     "sustainable": true,
-    "buffer_seconds": 60.355242856775796,
+    "buffer_seconds": 60.51814878517584,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347277.11690652283,
-    "upload_bytes_per_second": 3914233.5015917863,
+    "generation_bytes_per_second": 306154.79191120097,
+    "upload_bytes_per_second": 4109498.282917744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.51758509060635,
+    "estimated_full_shards_per_hour": 65.69369142534276,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 2545582,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.8264212399990356,
+      "bytes": 16670948,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 807500, "loss": 0.5008872747421265, "lr": 3e-05, "grad_norm": 0.26396072337872695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118194447543, "tokens_per_second": 150078.94842320183, "data_wait_s": 1.4904000636306591e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 807600, "loss": 0.5238015055656433, "lr": 3e-05, "grad_norm": 0.274179345818706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118209036743, "tokens_per_second": 151056.4577194445, "data_wait_s": 1.3855000361218117e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 807700, "loss": 0.5137314796447754, "lr": 3e-05, "grad_norm": 0.27360522484892497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118223849862, "tokens_per_second": 151407.2176996905, "data_wait_s": 1.4239998563425615e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 807800, "loss": 0.44607335329055786, "lr": 3e-05, "grad_norm": 0.316247497676291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118238535164, "tokens_per_second": 150838.82254082512, "data_wait_s": 1.5902000086498447e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f130de7d8e65ba013901e5d7739d7a088b05faa98e98ba8891624a707a9c7c8f
-size 164034
+oid sha256:14c8949000d1a6c1fecfd5c4e87cc83ec855b96c5a99b4201260ea5b59f54e9c
+size 166152