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": 76747.44719569046,
-      "required_pages_per_second_upper_estimate": 0.005309566475609805
+      "required_token_positions_per_second": 76476.41139722221,
+      "required_pages_per_second_upper_estimate": 0.005290815590182075
     },
     {
       "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": 9593.430899461307,
-      "required_pages_per_second_upper_estimate": 0.004016569183986574
+      "required_token_positions_per_second": 9559.551424652776,
+      "required_pages_per_second_upper_estimate": 0.004002384555368126
     },
     {
       "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": 19186.861798922615,
-      "required_pages_per_second_upper_estimate": 0.003996117321143484
+      "required_token_positions_per_second": 19119.102849305553,
+      "required_pages_per_second_upper_estimate": 0.003982004918861916
     },
     {
       "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": 28780.29269838392,
-      "required_pages_per_second_upper_estimate": 0.0017253346023056615
+      "required_token_positions_per_second": 28678.654273958327,
+      "required_pages_per_second_upper_estimate": 0.0017192415339542353
     },
     {
       "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": 28780.29269838392,
-      "required_pages_per_second_upper_estimate": 0.0022375411661703327
+      "required_token_positions_per_second": 28678.654273958327,
+      "required_pages_per_second_upper_estimate": 0.002229639225731425
     },
     {
       "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": 28780.29269838392,
-      "required_pages_per_second_upper_estimate": 0.36916269286417464
+      "required_token_positions_per_second": 28678.654273958327,
+      "required_pages_per_second_upper_estimate": 0.3678589842864808
     }
   ],
   "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": 0.08158012057036501,
-    "producer_p95_seconds": 0.2159172519986896,
-    "gpu_update_seconds": 0.9606573598729418,
-    "producer_batches_per_second": 12.257888233169178,
-    "gpu_batches_per_second": 1.0409538736394648,
-    "producer_headroom": 11.7756305476939,
+    "producer_batch_seconds": 0.08880520189846663,
+    "producer_p95_seconds": 0.1874910499991529,
+    "gpu_update_seconds": 0.9640619722211229,
+    "producer_batches_per_second": 11.26060161592028,
+    "gpu_batches_per_second": 1.0372777153486086,
+    "producer_headroom": 10.855917802240468,
     "sustainable": true,
-    "buffer_seconds": 60.52141367199533,
+    "buffer_seconds": 60.73590424993074,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317375.0233636038,
-    "upload_bytes_per_second": 4133040.138562268,
+    "generation_bytes_per_second": 337362.524992241,
+    "upload_bytes_per_second": 4081802.9223013143,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.10129190140806,
+    "estimated_full_shards_per_hour": 72.39014446568892,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 6380319,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1554255430000921,
+      "bytes": 1163617,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -92,3 +92,4 @@
 {"step": 809200, "loss": 0.5278820395469666, "lr": 3e-05, "grad_norm": 0.2742863426345246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118443379648, "tokens_per_second": 152686.85749597132, "data_wait_s": 9.901999874273315e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 809300, "loss": 0.537666916847229, "lr": 3e-05, "grad_norm": 0.26669529174821927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118457904094, "tokens_per_second": 149873.11519992395, "data_wait_s": 1.1402000382076949e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809400, "loss": 0.5624492168426514, "lr": 3e-05, "grad_norm": 0.27627468464037896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118472671545, "tokens_per_second": 152518.71777550018, "data_wait_s": 1.6554999092477374e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 809500, "loss": 0.5444309711456299, "lr": 3e-05, "grad_norm": 0.2800186850872327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118487430031, "tokens_per_second": 151739.90095548856, "data_wait_s": 1.8564998754300177e-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 sha256:87d3e72bf41af80ee32320f614e29bd95a55d709c6a8122e34d5abaaacf24fbd
-size 200040
+oid sha256:b5f2a2bdc644de73deab0fe132ae889bbaf0d2afda8f7660ad32b94aaaa99e5d
+size 202158