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": 74395.14592815646,
-      "required_pages_per_second_upper_estimate": 0.00644838269627908
+      "required_token_positions_per_second": 74743.92862263237,
+      "required_pages_per_second_upper_estimate": 0.006478614296254595
     },
     {
       "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": 9299.393241019558,
-      "required_pages_per_second_upper_estimate": 0.0037096670021619423
+      "required_token_positions_per_second": 9342.991077829047,
+      "required_pages_per_second_upper_estimate": 0.003727058831110997
     },
     {
       "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": 18598.786482039115,
-      "required_pages_per_second_upper_estimate": 0.004302501925157286
+      "required_token_positions_per_second": 18685.982155658094,
+      "required_pages_per_second_upper_estimate": 0.004322673109657595
     },
     {
       "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": 27898.17972305867,
-      "required_pages_per_second_upper_estimate": 0.0016988527746973697
+      "required_token_positions_per_second": 28028.97323348714,
+      "required_pages_per_second_upper_estimate": 0.0017068174132619481
     },
     {
       "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": 27898.17972305867,
-      "required_pages_per_second_upper_estimate": 0.0021476791021852845
+      "required_token_positions_per_second": 28028.97323348714,
+      "required_pages_per_second_upper_estimate": 0.002157747948677678
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27898.17972305867,
-      "required_pages_per_second_upper_estimate": 0.3464020229591203
+      "required_token_positions_per_second": 28028.97323348714,
+      "required_pages_per_second_upper_estimate": 0.3480260406209213
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.35602339974229835,
+    "producer_batch_seconds": 0.22077895186714613,
     "producer_p95_seconds": 0.3753280300006736,
-    "gpu_update_seconds": 0.9910323997643513,
-    "producer_batches_per_second": 2.808804142435114,
-    "gpu_batches_per_second": 1.0090487457703512,
-    "producer_headroom": 2.7836159097455218,
+    "gpu_update_seconds": 0.9864078776516872,
+    "producer_batches_per_second": 4.529417281597344,
+    "gpu_batches_per_second": 1.0137794138269365,
+    "producer_headroom": 4.467852887739311,
     "sustainable": true,
-    "buffer_seconds": 60.45297638562543,
+    "buffer_seconds": 60.170880536752925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438697.16514797084,
-    "upload_bytes_per_second": 3770826.8883538716,
+    "generation_bytes_per_second": 438797.84114155767,
+    "upload_bytes_per_second": 3727170.468716146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.13419929341644,
+    "estimated_full_shards_per_hour": 94.155802018023,
     "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": 0.9077771720003511,
-      "bytes": 6012024,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4875493239997013,
       "bytes": 6032602,
@@ -562,6 +556,12 @@
       "bytes": 10973458,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.8646178900016821,
+      "bytes": 947338,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -372,3 +372,4 @@
 {"step": 37200, "loss": 0.5847558341920376, "lr": 0.0003, "grad_norm": 0.123030510387659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5443579134, "tokens_per_second": 147852.19755474347, "data_wait_s": 1.693800004431978e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37300, "loss": 0.6867059618234634, "lr": 0.0003, "grad_norm": 0.14754163088994252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5458112551, "tokens_per_second": 147768.81894068664, "data_wait_s": 9.859999408945441e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 37400, "loss": 0.6461014598608017, "lr": 0.0003, "grad_norm": 0.131527074108428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5472766991, "tokens_per_second": 148598.9542381758, "data_wait_s": 8.20499553810805e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 37500, "loss": 0.6066446155309677, "lr": 0.0003, "grad_norm": 0.1288046809234849, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5487480301, "tokens_per_second": 149562.4722017392, "data_wait_s": 9.538998710922897e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6533be99ab3387f6121df33b92567edd75c77c348b863f719c2a1589bc1c2b5
-size 792004
+oid sha256:75f31d77723c3b0e799b967de24c1ded6b93d3cc30ccc2fb12bfd6fe892a5af8
+size 794122