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": 76596.26997431864,
-      "required_pages_per_second_upper_estimate": 0.005299107684655821
+      "required_token_positions_per_second": 75844.1368570214,
+      "required_pages_per_second_upper_estimate": 0.005247073370411925
     },
     {
       "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": 9574.53374678983,
-      "required_pages_per_second_upper_estimate": 0.004008657340780447
+      "required_token_positions_per_second": 9480.517107127675,
+      "required_pages_per_second_upper_estimate": 0.003969294537044592
     },
     {
       "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": 19149.06749357966,
-      "required_pages_per_second_upper_estimate": 0.003988245764043404
+      "required_token_positions_per_second": 18961.03421425535,
+      "required_pages_per_second_upper_estimate": 0.003949083390731188
     },
     {
       "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": 28723.601240369488,
-      "required_pages_per_second_upper_estimate": 0.0017219360359605404
+      "required_token_positions_per_second": 28441.551321383024,
+      "required_pages_per_second_upper_estimate": 0.001705027573982592
     },
     {
       "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": 28723.601240369488,
-      "required_pages_per_second_upper_estimate": 0.002233133654669081
+      "required_token_positions_per_second": 28441.551321383024,
+      "required_pages_per_second_upper_estimate": 0.002211205514074366
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28723.601240369488,
-      "required_pages_per_second_upper_estimate": 0.3684355157113106
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28441.551321383024,
+      "required_pages_per_second_upper_estimate": 0.3648176821921605
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11593528183590252,
-    "producer_p95_seconds": 0.34982289600156946,
-    "gpu_update_seconds": 0.962553398810669,
-    "producer_batches_per_second": 8.625501953886852,
-    "gpu_batches_per_second": 1.0389034013443825,
-    "producer_headroom": 8.302506222161856,
+    "producer_batch_seconds": 0.4029591803591188,
+    "producer_p95_seconds": 0.3505544009967707,
+    "gpu_update_seconds": 0.9720988734961721,
+    "producer_batches_per_second": 2.4816409421639087,
+    "gpu_batches_per_second": 1.0287019430477076,
+    "producer_headroom": 2.412400364299515,
     "sustainable": true,
-    "buffer_seconds": 60.64086412507215,
+    "buffer_seconds": 60.27013015676267,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352989.70656987414,
-    "upload_bytes_per_second": 4160110.949392872,
+    "generation_bytes_per_second": 324202.0926816801,
+    "upload_bytes_per_second": 4184247.457970663,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.74337385008019,
+    "estimated_full_shards_per_hour": 69.56622205102732,
     "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": 1.6914732460027153,
-      "bytes": 2366813,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.8487111699978414,
       "bytes": 10763047,
@@ -562,6 +556,12 @@
       "bytes": 13513556,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.8802343610004755,
+      "bytes": 5656226,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -392,3 +392,4 @@
 {"step": 839200, "loss": 0.5457845330238342, "lr": 3e-05, "grad_norm": 0.28252370980252123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122835090976, "tokens_per_second": 143979.019803392, "data_wait_s": 1.4349003322422504e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 839300, "loss": 0.546668291091919, "lr": 3e-05, "grad_norm": 0.3009325826473563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122849542127, "tokens_per_second": 149136.20504048473, "data_wait_s": 9.843999578151852e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 839400, "loss": 0.576280415058136, "lr": 3e-05, "grad_norm": 0.3093915645088455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122864220687, "tokens_per_second": 151340.57215926223, "data_wait_s": 1.3443997886497527e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 839500, "loss": 0.5308698415756226, "lr": 3e-05, "grad_norm": 0.28809007191134933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122878902505, "tokens_per_second": 151331.53464158325, "data_wait_s": 1.705899921944365e-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 sha256:61edc64f86daf28813acd59371bd03c71adde54f169e03a53e816be1659be7b7
-size 840138
+oid sha256:4882e9b3abad1044014bedde80c5aae4896ac16bbac4fdddffcecb6228cfb2d8
+size 842256