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": 75378.60883438532,
-      "required_pages_per_second_upper_estimate": 0.00653362676842705
+      "required_token_positions_per_second": 75293.58784299363,
+      "required_pages_per_second_upper_estimate": 0.006526257364376951
     },
     {
       "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": 9422.326104298165,
-      "required_pages_per_second_upper_estimate": 0.003758706759333878
+      "required_token_positions_per_second": 9411.698480374203,
+      "required_pages_per_second_upper_estimate": 0.003754467241253472
     },
     {
       "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": 18844.65220859633,
-      "required_pages_per_second_upper_estimate": 0.004359378633907299
+      "required_token_positions_per_second": 18823.396960748407,
+      "required_pages_per_second_upper_estimate": 0.004354461606397268
     },
     {
       "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": 28266.97831289449,
-      "required_pages_per_second_upper_estimate": 0.00172131067388171
+      "required_token_positions_per_second": 28235.095441122612,
+      "required_pages_per_second_upper_estimate": 0.001719369174267303
     },
     {
       "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": 28266.97831289449,
-      "required_pages_per_second_upper_estimate": 0.002176070238530683
+      "required_token_positions_per_second": 28235.095441122612,
+      "required_pages_per_second_upper_estimate": 0.002173615806804246
     },
     {
       "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": 28266.97831289449,
-      "required_pages_per_second_upper_estimate": 0.3509812671554005
+      "required_token_positions_per_second": 28235.095441122612,
+      "required_pages_per_second_upper_estimate": 0.35058538859310145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5108880850780508,
-    "producer_p95_seconds": 0.24243838199981838,
-    "gpu_update_seconds": 0.9781024237524485,
-    "producer_batches_per_second": 1.9573758504217713,
-    "gpu_batches_per_second": 1.0223878151365196,
-    "producer_headroom": 1.9145140634920446,
+    "producer_batch_seconds": 0.2817690258203811,
+    "producer_p95_seconds": 0.25606173199776094,
+    "gpu_update_seconds": 0.9792068901503501,
+    "producer_batches_per_second": 3.5490061304235354,
+    "gpu_batches_per_second": 1.021234644137826,
+    "producer_headroom": 3.475211256096558,
     "sustainable": true,
-    "buffer_seconds": 60.642350272651804,
+    "buffer_seconds": 60.710827189321705,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 201632.29767382308,
-    "upload_bytes_per_second": 6209025.938223994,
+    "generation_bytes_per_second": 201050.54205535655,
+    "upload_bytes_per_second": 6139180.410517605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 43.265597321138564,
+    "estimated_full_shards_per_hour": 43.140766107993336,
     "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": 0.9871070610006427,
-      "bytes": 12008396,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 1.228597503999481,
       "bytes": 13077636,
@@ -562,6 +556,12 @@
       "bytes": 752773,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.0778829359987867,
+      "bytes": 7678507,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -162,3 +162,4 @@
 {"step": 16200, "loss": 0.7200316488742828, "lr": 0.0003, "grad_norm": 0.14547465273479035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2370236048, "tokens_per_second": 147838.81526074515, "data_wait_s": 1.0864998330362141e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 16300, "loss": 0.6719108186662197, "lr": 0.0003, "grad_norm": 0.13416477956147393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2384862018, "tokens_per_second": 148957.66477971146, "data_wait_s": 1.0330997611163184e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 16400, "loss": 0.6719970945268869, "lr": 0.0003, "grad_norm": 0.15009249968606697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2399525248, "tokens_per_second": 142396.41128679027, "data_wait_s": 1.2108997907489538e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 16500, "loss": 0.6644943132996559, "lr": 0.0003, "grad_norm": 0.14038254831626634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2414226956, "tokens_per_second": 149425.55997097335, "data_wait_s": 9.78800017037429e-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 sha256:98d33c42fe98df4161ecbf77eb8183b06adb729a4a88f729a296d67c19755595
-size 344092
+oid sha256:e564d62901489f71af3d714fa51f69c0130b5635deaa628869bf5320a220708d
+size 346210