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": 75777.8227057171,
-      "required_pages_per_second_upper_estimate": 0.006568229614995783
+      "required_token_positions_per_second": 76093.40616834645,
+      "required_pages_per_second_upper_estimate": 0.006595583589697521
     },
     {
       "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": 9472.227838214638,
-      "required_pages_per_second_upper_estimate": 0.003778613307090569
+      "required_token_positions_per_second": 9511.675771043307,
+      "required_pages_per_second_upper_estimate": 0.00379434967729508
     },
     {
       "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": 18944.455676429276,
-      "required_pages_per_second_upper_estimate": 0.0043824664097094086
+      "required_token_positions_per_second": 19023.351542086613,
+      "required_pages_per_second_upper_estimate": 0.004400717579709422
     },
     {
       "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": 28416.683514643915,
-      "required_pages_per_second_upper_estimate": 0.0017304269352257586
+      "required_token_positions_per_second": 28535.02731312992,
+      "required_pages_per_second_upper_estimate": 0.0017376334516516312
     },
     {
       "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": 28416.683514643915,
-      "required_pages_per_second_upper_estimate": 0.0021875949593718755
+      "required_token_positions_per_second": 28535.02731312992,
+      "required_pages_per_second_upper_estimate": 0.0021967053925759294
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28416.683514643915,
-      "required_pages_per_second_upper_estimate": 0.3528401047300485
+      "required_token_positions_per_second": 28535.02731312992,
+      "required_pages_per_second_upper_estimate": 0.35430953863602965
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.4002482995624064,
-    "producer_p95_seconds": 0.3779312479964574,
-    "gpu_update_seconds": 0.9729495697748194,
-    "producer_batches_per_second": 2.49844909046036,
-    "gpu_batches_per_second": 1.0278024998062758,
-    "producer_headroom": 2.4308649676676963,
+    "producer_batch_seconds": 0.1169289658594721,
+    "producer_p95_seconds": 0.3576279959961539,
+    "gpu_update_seconds": 0.9689144396675672,
+    "producer_batches_per_second": 8.552200839626195,
+    "gpu_batches_per_second": 1.0320828744621644,
+    "producer_headroom": 8.286350884450913,
     "sustainable": true,
-    "buffer_seconds": 60.3228733260388,
+    "buffer_seconds": 60.072695259389164,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313997.160096495,
-    "upload_bytes_per_second": 5140850.885270301,
+    "generation_bytes_per_second": 303009.0750415751,
+    "upload_bytes_per_second": 5205629.805969638,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.37648107691896,
+    "estimated_full_shards_per_hour": 65.01869381366195,
     "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": 1.3764796949981246,
-      "bytes": 2478260,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1128817630014964,
       "bytes": 6162674,
@@ -562,6 +556,12 @@
       "bytes": 4758637,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.040413394002826,
+      "bytes": 5884125,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 272200, "loss": 0.5812223106622696, "lr": 0.0003, "grad_norm": 0.1554620372090776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39841620020, "tokens_per_second": 149822.57141347372, "data_wait_s": 1.4734003343619406e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272300, "loss": 0.5821655839681625, "lr": 0.0003, "grad_norm": 0.15577968835783318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39856249142, "tokens_per_second": 148881.06254462423, "data_wait_s": 1.9008999515790492e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272400, "loss": 0.6329742148518562, "lr": 0.0003, "grad_norm": 0.142750450724264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39870833835, "tokens_per_second": 148554.5394107169, "data_wait_s": 1.657800021348521e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 272500, "loss": 0.5857648644596338, "lr": 0.0003, "grad_norm": 0.16595507092760214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39885375281, "tokens_per_second": 136249.45000694657, "data_wait_s": 1.1500997061375529e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e3f632da24a03d6221b4dda549ca1b37b63a39f22674e474b5cb14a86429757
-size 753300
+oid sha256:7516ca12c3d00e230180c0a30288e398ec6e95548f910d4e3cecbe2132c197a3
+size 755418