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": 74351.25676561083,
-      "required_pages_per_second_upper_estimate": 0.0064445784949056905
+      "required_token_positions_per_second": 74166.92089029694,
+      "required_pages_per_second_upper_estimate": 0.006428600728428487
     },
     {
       "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": 9293.907095701354,
-      "required_pages_per_second_upper_estimate": 0.003707478496769329
+      "required_token_positions_per_second": 9270.865111287118,
+      "required_pages_per_second_upper_estimate": 0.0036982867046781226
     },
     {
       "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": 18587.814191402707,
-      "required_pages_per_second_upper_estimate": 0.004299963678824273
+      "required_token_positions_per_second": 18541.730222574235,
+      "required_pages_per_second_upper_estimate": 0.004289302963685957
     },
     {
       "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": 27881.72128710406,
-      "required_pages_per_second_upper_estimate": 0.0016978505422984738
+      "required_token_positions_per_second": 27812.59533386135,
+      "required_pages_per_second_upper_estimate": 0.0016936411344218403
     },
     {
       "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": 27881.72128710406,
-      "required_pages_per_second_upper_estimate": 0.002146412086225632
+      "required_token_positions_per_second": 27812.59533386135,
+      "required_pages_per_second_upper_estimate": 0.0021410905789921225
     },
     {
       "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": 27881.72128710406,
-      "required_pages_per_second_upper_estimate": 0.3461976642674058
+      "required_token_positions_per_second": 27812.59533386135,
+      "required_pages_per_second_upper_estimate": 0.34533935127781457
     }
   ],
   "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": 0.15568664225003204,
+    "producer_batch_seconds": 0.12914335346073358,
     "producer_p95_seconds": 0.4492218699961086,
-    "gpu_update_seconds": 0.9916174010672661,
-    "producer_batches_per_second": 6.423158631644227,
-    "gpu_batches_per_second": 1.0084534609050948,
-    "producer_headroom": 6.369315868953826,
+    "gpu_update_seconds": 0.9940819858094667,
+    "producer_batches_per_second": 7.743333072917709,
+    "gpu_batches_per_second": 1.0059532455823694,
+    "producer_headroom": 7.697507917910157,
     "sustainable": true,
-    "buffer_seconds": 60.488661465103235,
+    "buffer_seconds": 60.63900113437747,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331426.88358465076,
-    "upload_bytes_per_second": 3690108.4158990155,
+    "generation_bytes_per_second": 378784.97178537335,
+    "upload_bytes_per_second": 3686780.616802958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.11649399428026,
+    "estimated_full_shards_per_hour": 81.27843728228474,
     "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": 1.014886224998918,
-      "bytes": 6118927,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9393350060054217,
       "bytes": 807847,
@@ -562,6 +556,12 @@
       "bytes": 8317362,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.006012271005602,
+      "bytes": 5704285,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -404,3 +404,4 @@
 {"step": 40400, "loss": 0.6147771067917347, "lr": 0.0003, "grad_norm": 0.1216657546652096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5911786558, "tokens_per_second": 140943.46984423554, "data_wait_s": 1.259699638467282e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40500, "loss": 0.6718204841017723, "lr": 0.0003, "grad_norm": 0.14208304979168693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5926359613, "tokens_per_second": 147473.17327324444, "data_wait_s": 1.3800003216601908e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40600, "loss": 0.6331306956708431, "lr": 0.0003, "grad_norm": 0.12806531454651754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5941002069, "tokens_per_second": 147839.9956131124, "data_wait_s": 1.1591000657062978e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 40700, "loss": 0.6462833769619465, "lr": 0.0003, "grad_norm": 0.12658802811712694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5955780384, "tokens_per_second": 149595.9284136709, "data_wait_s": 1.1096999514847994e-05, "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 sha256:1e89cce787ddb301e1fc1295cb5a11f9f1453ac2b9d48c88c29d3b45d523fb65
-size 860563
+oid sha256:508e5e3996e733c414f0d648c139decac6bf277547def1c621bcb985b778bfe0
+size 862681