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": 76707.98802974932,
-      "required_pages_per_second_upper_estimate": 0.005306836598952118
+      "required_token_positions_per_second": 76703.28817277803,
+      "required_pages_per_second_upper_estimate": 0.005306511451941673
     },
     {
       "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": 9588.498503718665,
-      "required_pages_per_second_upper_estimate": 0.004014504092889265
+      "required_token_positions_per_second": 9587.911021597254,
+      "required_pages_per_second_upper_estimate": 0.004014258126393052
     },
     {
       "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": 19176.99700743733,
-      "required_pages_per_second_upper_estimate": 0.0039940627452291445
+      "required_token_positions_per_second": 19175.82204319451,
+      "required_pages_per_second_upper_estimate": 0.003993818031163256
     },
     {
       "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": 28765.495511155994,
-      "required_pages_per_second_upper_estimate": 0.001724447533525344
+      "required_token_positions_per_second": 28763.733064791766,
+      "required_pages_per_second_upper_estimate": 0.0017243418775569083
     },
     {
       "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": 28765.495511155994,
-      "required_pages_per_second_upper_estimate": 0.0022363907499493092
+      "required_token_positions_per_second": 28763.733064791766,
+      "required_pages_per_second_upper_estimate": 0.002236253727496673
     },
     {
       "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": 28765.495511155994,
-      "required_pages_per_second_upper_estimate": 0.3689728904343966
+      "required_token_positions_per_second": 28763.733064791766,
+      "required_pages_per_second_upper_estimate": 0.36895028366480376
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0898498250781472,
-    "producer_p95_seconds": 0.25922722299947054,
-    "gpu_update_seconds": 0.9611515292436871,
-    "producer_batches_per_second": 11.129682212850682,
-    "gpu_batches_per_second": 1.04041867444864,
-    "producer_headroom": 10.697311078877695,
+    "producer_batch_seconds": 0.0980609001722712,
+    "producer_p95_seconds": 0.2802222360005544,
+    "gpu_update_seconds": 0.9612104220867813,
+    "producer_batches_per_second": 10.19774444496453,
+    "gpu_batches_per_second": 1.04035492855873,
+    "producer_headroom": 9.802178242277485,
     "sustainable": true,
-    "buffer_seconds": 60.55254634235229,
+    "buffer_seconds": 60.556256591467225,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386779.91067480634,
-    "upload_bytes_per_second": 6126065.34538016,
+    "generation_bytes_per_second": 372493.93079099775,
+    "upload_bytes_per_second": 6101851.120600281,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.99396505530493,
+    "estimated_full_shards_per_hour": 79.9285263328309,
     "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": 1.054139716001373,
-      "bytes": 8196073,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1359631509985775,
       "bytes": 6053377,
@@ -562,6 +556,12 @@
       "bytes": 2572134,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.036350621998281,
+      "bytes": 6380141,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 509200, "loss": 0.5866506695747375, "lr": 0.0003, "grad_norm": 0.15648726396603002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74529486352, "tokens_per_second": 149619.91736748046, "data_wait_s": 1.5210000128718093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509300, "loss": 0.5844889879226685, "lr": 0.0003, "grad_norm": 0.16537971772199328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74543980505, "tokens_per_second": 150588.78921485308, "data_wait_s": 1.4104000001680106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509400, "loss": 0.6208956241607666, "lr": 0.0003, "grad_norm": 0.17751308819122566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74558482958, "tokens_per_second": 150978.5163113281, "data_wait_s": 2.1413001377368346e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 509500, "loss": 0.735479474067688, "lr": 0.0003, "grad_norm": 0.21468405559765869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74573229358, "tokens_per_second": 153840.5896974438, "data_wait_s": 1.0335003025829792e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d2b856f17de8d4be5107fd36ebfbfecd942345aa7809ca9a67d7c7543932145
-size 520089
+oid sha256:9a90d863470b8bd10a5b696b81c93730c0d9bd6d99de91acee0498d6f5abd61f
+size 522207