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": 76559.38087740359,
-      "required_pages_per_second_upper_estimate": 0.00529655561133674
+      "required_token_positions_per_second": 76647.86940891923,
+      "required_pages_per_second_upper_estimate": 0.005302677453268673
     },
     {
       "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": 9569.922609675448,
-      "required_pages_per_second_upper_estimate": 0.004006726753962148
+      "required_token_positions_per_second": 9580.983676114904,
+      "required_pages_per_second_upper_estimate": 0.004011357791499015
     },
     {
       "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": 19139.845219350897,
-      "required_pages_per_second_upper_estimate": 0.003986325007529278
+      "required_token_positions_per_second": 19161.967352229807,
+      "required_pages_per_second_upper_estimate": 0.00399093246440808
     },
     {
       "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": 28709.767829026347,
-      "required_pages_per_second_upper_estimate": 0.0017211067440729153
+      "required_token_positions_per_second": 28742.95102834471,
+      "required_pages_per_second_upper_estimate": 0.0017230960262042394
     },
     {
       "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": 28709.767829026347,
-      "required_pages_per_second_upper_estimate": 0.0022320581677838978
+      "required_token_positions_per_second": 28742.95102834471,
+      "required_pages_per_second_upper_estimate": 0.002234638015573432
     },
     {
       "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": 28709.767829026347,
-      "required_pages_per_second_upper_estimate": 0.36825807556376067
+      "required_token_positions_per_second": 28742.95102834471,
+      "required_pages_per_second_upper_estimate": 0.3686837140152731
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.27156292776558644,
-    "producer_p95_seconds": 0.1719944240003315,
-    "gpu_update_seconds": 0.9630171920807779,
-    "producer_batches_per_second": 3.6823877553095232,
-    "gpu_batches_per_second": 1.0384030609456867,
-    "producer_headroom": 3.546202716270815,
+    "producer_batch_seconds": 0.04623263098437036,
+    "producer_p95_seconds": 0.1578233509999336,
+    "gpu_update_seconds": 0.9619054067459903,
+    "producer_batches_per_second": 21.629744591824444,
+    "gpu_batches_per_second": 1.039603263467329,
+    "producer_headroom": 20.805768269410777,
     "sustainable": true,
-    "buffer_seconds": 60.670083101089006,
+    "buffer_seconds": 60.60004062499739,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 512938.7331526132,
-    "upload_bytes_per_second": 4248713.488359095,
+    "generation_bytes_per_second": 498007.38767096004,
+    "upload_bytes_per_second": 4260307.658308167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.06471153196142,
+    "estimated_full_shards_per_hour": 106.86079237553216,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 12020074,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.294827431000158,
+      "bytes": 6143505,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 846700, "loss": 0.4852267801761627, "lr": 3e-05, "grad_norm": 0.2792860797559969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123933412034, "tokens_per_second": 152542.14170023164, "data_wait_s": 2.0462000065890606e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846800, "loss": 0.5162094831466675, "lr": 3e-05, "grad_norm": 0.29515991358143606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123948088795, "tokens_per_second": 153339.32341996324, "data_wait_s": 8.813000022200868e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 846900, "loss": 0.5729007124900818, "lr": 3e-05, "grad_norm": 0.2864219809632993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123962790066, "tokens_per_second": 152166.5793992089, "data_wait_s": 1.4643999747931957e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 847000, "loss": 0.5024833083152771, "lr": 3e-05, "grad_norm": 0.2924154492353622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123977571845, "tokens_per_second": 153906.670506806, "data_wait_s": 1.4862000170978718e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf06ed06ad2e35d5df533f300f69a5bd48fb4c65a48fa2a668f7c8b9bd7138dc
-size 94140
+oid sha256:480d60269a08ac2c5e3a84bdc5c21fb9c9c4b6597f6f7f12cd9de690337619bd
+size 96258