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": 75333.92354336802,
-      "required_pages_per_second_upper_estimate": 0.006529753560655505
+      "required_token_positions_per_second": 75398.22431468604,
+      "required_pages_per_second_upper_estimate": 0.006535326988544536
     },
     {
       "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": 9416.740442921002,
-      "required_pages_per_second_upper_estimate": 0.0037564785554974477
+      "required_token_positions_per_second": 9424.778039335755,
+      "required_pages_per_second_upper_estimate": 0.0037596848728800683
     },
     {
       "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": 18833.480885842004,
-      "required_pages_per_second_upper_estimate": 0.004356794344996651
+      "required_token_positions_per_second": 18849.55607867151,
+      "required_pages_per_second_upper_estimate": 0.004360513057944026
     },
     {
       "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": 28250.221328763004,
-      "required_pages_per_second_upper_estimate": 0.0017202902614652034
+      "required_token_positions_per_second": 28274.334118007264,
+      "required_pages_per_second_upper_estimate": 0.001721758603820151
     },
     {
       "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": 28250.221328763004,
-      "required_pages_per_second_upper_estimate": 0.00217478023950594
+      "required_token_positions_per_second": 28274.334118007264,
+      "required_pages_per_second_upper_estimate": 0.002176636508770436
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28250.221328763004,
-      "required_pages_per_second_upper_estimate": 0.3507732014945057
+      "required_token_positions_per_second": 28274.334118007264,
+      "required_pages_per_second_upper_estimate": 0.3510726016366051
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10580215948442628,
-    "producer_p95_seconds": 0.4161550639983034,
-    "gpu_update_seconds": 0.9786825978545572,
-    "producer_batches_per_second": 9.45160292448659,
-    "gpu_batches_per_second": 1.0217817320877822,
-    "producer_headroom": 9.250119304026265,
+    "producer_batch_seconds": 0.2636677381327104,
+    "producer_p95_seconds": 0.42771556099614827,
+    "gpu_update_seconds": 0.9778479622051164,
+    "producer_batches_per_second": 3.7926520972265316,
+    "gpu_batches_per_second": 1.0226538671154248,
+    "producer_headroom": 3.708637124625925,
     "sustainable": true,
-    "buffer_seconds": 60.67832106698255,
+    "buffer_seconds": 60.62657365671721,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "upload_pages_per_commit": 155,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 611318.1893232007,
+    "generation_bytes_per_second": 591876.4461234412,
     "upload_bytes_per_second": 5834285.3601310225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.58732633481988,
+    "estimated_full_shards_per_hour": 63.50145357979501,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -409,3 +409,4 @@
 {"step": 155900, "loss": 0.5951719842851162, "lr": 0.0003, "grad_norm": 0.15011265849159988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22815968278, "tokens_per_second": 148507.44070253478, "data_wait_s": 1.1504002031870186e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156000, "loss": 0.6793359220027924, "lr": 0.0003, "grad_norm": 0.14165331519268554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22830648258, "tokens_per_second": 148951.80534687472, "data_wait_s": 1.4250996173359454e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156100, "loss": 0.5843771025538445, "lr": 0.0003, "grad_norm": 0.13198605677032757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22845328580, "tokens_per_second": 149686.82104776063, "data_wait_s": 1.4117002137936652e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 156200, "loss": 0.5766946114599705, "lr": 0.0003, "grad_norm": 0.12941958180573832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22859969626, "tokens_per_second": 149426.02031784237, "data_wait_s": 1.3151999155525118e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b280d3d32cc7999e1f71c1e57e681f286833e976f6f976dc36a8b2ec3e44fe0
-size 876927
+oid sha256:b1fdb6aa3b572a44ec17c899c12415af9d41b0afe95b7444906e9775859fcbda
+size 879045