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": 76916.13638660368,
-      "required_pages_per_second_upper_estimate": 0.005321236785250031
+      "required_token_positions_per_second": 76692.75704011027,
+      "required_pages_per_second_upper_estimate": 0.0053057828837481235
     },
     {
       "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": 9614.51704832546,
-      "required_pages_per_second_upper_estimate": 0.004025397514187135
+      "required_token_positions_per_second": 9586.594630013784,
+      "required_pages_per_second_upper_estimate": 0.004013706980726435
     },
     {
       "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": 19229.03409665092,
-      "required_pages_per_second_upper_estimate": 0.004004900698602009
+      "required_token_positions_per_second": 19173.189260027568,
+      "required_pages_per_second_upper_estimate": 0.003993269691860743
     },
     {
       "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": 28843.551144976384,
-      "required_pages_per_second_upper_estimate": 0.0017291268495888245
+      "required_token_positions_per_second": 28759.78389004135,
+      "required_pages_per_second_upper_estimate": 0.0017241051305606615
     },
     {
       "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": 28843.551144976384,
-      "required_pages_per_second_upper_estimate": 0.0022424592321483956
+      "required_token_positions_per_second": 28759.78389004135,
+      "required_pages_per_second_upper_estimate": 0.0022359466965304123
     },
     {
       "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": 28843.551144976384,
-      "required_pages_per_second_upper_estimate": 0.3699741042954347
+      "required_token_positions_per_second": 28759.78389004135,
+      "required_pages_per_second_upper_estimate": 0.3688996278913989
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11683065746090904,
-    "producer_p95_seconds": 0.19626906999837956,
-    "gpu_update_seconds": 0.9585504871100239,
-    "producer_batches_per_second": 8.559397179927666,
-    "gpu_batches_per_second": 1.0432418672228148,
-    "producer_headroom": 8.204614336187829,
+    "producer_batch_seconds": 0.07588782031248797,
+    "producer_p95_seconds": 0.19041198800005077,
+    "gpu_update_seconds": 0.9613424115322949,
+    "producer_batches_per_second": 13.177345137628649,
+    "gpu_batches_per_second": 1.0402120909303152,
+    "producer_headroom": 12.667940752201286,
     "sustainable": true,
-    "buffer_seconds": 60.388680687931505,
+    "buffer_seconds": 60.56457192653458,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307488.8368280104,
-    "upload_bytes_per_second": 5460052.1133410875,
+    "generation_bytes_per_second": 309447.24291663105,
+    "upload_bytes_per_second": 5596286.624996297,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.9799464095138,
+    "estimated_full_shards_per_hour": 66.4001747667713,
     "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": 1.7620159190000777,
-      "bytes": 5407000,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.0307093159999567,
       "bytes": 7655434,
@@ -562,6 +556,12 @@
       "bytes": 6509260,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2692650000008143,
+      "bytes": 13237909,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 616900, "loss": 0.524788498878479, "lr": 0.0001, "grad_norm": 0.1643309147252453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90296100518, "tokens_per_second": 152226.6330800492, "data_wait_s": 1.485800021328032e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 617000, "loss": 0.5374523401260376, "lr": 0.0001, "grad_norm": 0.15429343033973236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90310856570, "tokens_per_second": 130200.13436249521, "data_wait_s": 1.4109999028733e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 617100, "loss": 0.5785061717033386, "lr": 0.0001, "grad_norm": 0.14414073993283827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90325585542, "tokens_per_second": 153308.45878248534, "data_wait_s": 1.254800008609891e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 617200, "loss": 0.5823729038238525, "lr": 0.0001, "grad_norm": 0.14502301136742657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90340333233, "tokens_per_second": 153932.43073792415, "data_wait_s": 9.142999260802753e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1369f1bf81990d2c0158b790999bba8239edfd18e79a8701aab88cf5e0d8502
-size 271500
+oid sha256:136b1611ad7306664b121d394315f76b6fd48c2c1001690560650110c4ea1072
+size 273618