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": 76586.55261667496,
-      "required_pages_per_second_upper_estimate": 0.005298435415306655
+      "required_token_positions_per_second": 76814.27046601212,
+      "required_pages_per_second_upper_estimate": 0.005314189464502007
     },
     {
       "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": 9573.31907708437,
-      "required_pages_per_second_upper_estimate": 0.004008148783939959
+      "required_token_positions_per_second": 9601.783808251515,
+      "required_pages_per_second_upper_estimate": 0.004020066372468463
     },
     {
       "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": 19146.63815416874,
-      "required_pages_per_second_upper_estimate": 0.003987739796710098
+      "required_token_positions_per_second": 19203.56761650303,
+      "required_pages_per_second_upper_estimate": 0.0039995967023834485
     },
     {
       "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": 28719.957231253113,
-      "required_pages_per_second_upper_estimate": 0.0017217175831780935
+      "required_token_positions_per_second": 28805.351424754546,
+      "required_pages_per_second_upper_estimate": 0.0017268368346891208
     },
     {
       "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": 28719.957231253113,
-      "required_pages_per_second_upper_estimate": 0.002232850348988585
+      "required_token_positions_per_second": 28805.351424754546,
+      "required_pages_per_second_upper_estimate": 0.00223948937192396
     },
     {
       "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": 28719.957231253113,
-      "required_pages_per_second_upper_estimate": 0.36838877427499794
+      "required_token_positions_per_second": 28805.351424754546,
+      "required_pages_per_second_upper_estimate": 0.36948411930009295
     }
   ],
   "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": 0.35227885468748354,
+    "producer_batch_seconds": 0.11532967796895832,
     "producer_p95_seconds": 0.24070170299819438,
-    "gpu_update_seconds": 0.9626755282878136,
-    "producer_batches_per_second": 2.8386603018995507,
-    "gpu_batches_per_second": 1.0387716012461339,
-    "producer_headroom": 2.7327088057607942,
+    "gpu_update_seconds": 0.9598216523142311,
+    "producer_batches_per_second": 8.670795042618224,
+    "gpu_batches_per_second": 1.0418602222495135,
+    "producer_headroom": 8.322416824683868,
     "sustainable": true,
-    "buffer_seconds": 60.64855828213226,
+    "buffer_seconds": 60.46876409579656,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309908.0002141861,
-    "upload_bytes_per_second": 5913692.744159812,
+    "generation_bytes_per_second": 322720.1593479705,
+    "upload_bytes_per_second": 5933281.998072939,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.49904255694568,
+    "estimated_full_shards_per_hour": 69.24823365525566,
     "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": 1.2017747059999238,
-      "bytes": 5823273,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.1922360469998239,
       "bytes": 10997251,
@@ -562,6 +556,12 @@
       "bytes": 5953132,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.9063561590010067,
+      "bytes": 11430108,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -146,3 +146,4 @@
 {"step": 499600, "loss": 0.5783843994140625, "lr": 0.0003, "grad_norm": 0.14725862291518543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73124620601, "tokens_per_second": 152956.6464081035, "data_wait_s": 1.386900112265721e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499700, "loss": 0.6527587175369263, "lr": 0.0003, "grad_norm": 0.22528530225703736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73139199520, "tokens_per_second": 151710.46077142513, "data_wait_s": 1.0285999451298267e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499800, "loss": 0.5949404239654541, "lr": 0.0003, "grad_norm": 0.16278332905207246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73153770130, "tokens_per_second": 151713.50425051744, "data_wait_s": 1.686999894445762e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 499900, "loss": 0.5922921299934387, "lr": 0.0003, "grad_norm": 0.1657469942759339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73168359778, "tokens_per_second": 152277.53785663372, "data_wait_s": 1.4310000551631674e-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:a986feb75f9b12fddf6c03fe5de0c1d170b1a1ec2a1dae557a9adc67c5c4a385
-size 315195
+oid sha256:85c62638ffffd497a29d9d04e16a1008ad3c5961e1b9051f7faed9ff1feab8e7
+size 317313