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": 75614.30079503529,
-      "required_pages_per_second_upper_estimate": 0.006554055950220377
+      "required_token_positions_per_second": 76180.71786411507,
+      "required_pages_per_second_upper_estimate": 0.006603151546197275
     },
     {
       "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": 9451.787599379411,
-      "required_pages_per_second_upper_estimate": 0.003770459390210392
+      "required_token_positions_per_second": 9522.589733014383,
+      "required_pages_per_second_upper_estimate": 0.0037987034199036154
     },
     {
       "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": 18903.575198758823,
-      "required_pages_per_second_upper_estimate": 0.004373009430672183
+      "required_token_positions_per_second": 19045.179466028767,
+      "required_pages_per_second_upper_estimate": 0.004405767085755098
     },
     {
       "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": 28355.362798138234,
-      "required_pages_per_second_upper_estimate": 0.0017266928253154977
+      "required_token_positions_per_second": 28567.76919904315,
+      "required_pages_per_second_upper_estimate": 0.0017396272607203478
     },
     {
       "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": 28355.362798138234,
-      "required_pages_per_second_upper_estimate": 0.0021828743208686606
+      "required_token_positions_per_second": 28567.76919904315,
+      "required_pages_per_second_upper_estimate": 0.002199225953588871
     },
     {
       "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": 28355.362798138234,
-      "required_pages_per_second_upper_estimate": 0.35207870665828417
+      "required_token_positions_per_second": 28567.76919904315,
+      "required_pages_per_second_upper_estimate": 0.35471608327902887
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10317348972699847,
-    "producer_p95_seconds": 0.42441760600195266,
-    "gpu_update_seconds": 0.9750536502327991,
-    "producer_batches_per_second": 9.692412291626884,
-    "gpu_batches_per_second": 1.0255845919465507,
-    "producer_headroom": 9.450621984512043,
+    "producer_batch_seconds": 0.16833199735197013,
+    "producer_p95_seconds": 0.43461761200160254,
+    "gpu_update_seconds": 0.9678039544272867,
+    "producer_batches_per_second": 5.940641207441219,
+    "gpu_batches_per_second": 1.0332671151274286,
+    "producer_headroom": 5.749376052395304,
     "sustainable": true,
-    "buffer_seconds": 60.453326314433546,
+    "buffer_seconds": 60.00384517449178,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298429.03024260147,
-    "upload_bytes_per_second": 6317206.668870772,
+    "generation_bytes_per_second": 310285.52020252333,
+    "upload_bytes_per_second": 6264055.424540229,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.03592281778845,
+    "estimated_full_shards_per_hour": 66.5800495582273,
     "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": 1.0164505040011136,
-      "bytes": 6233089,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.140030457998364,
       "bytes": 2525214,
@@ -562,6 +556,12 @@
       "bytes": 6150922,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.967503943000338,
+      "bytes": 2353719,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 79300, "loss": 0.605455256998539, "lr": 0.0003, "grad_norm": 0.13833420435603655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11605053680, "tokens_per_second": 149502.2979602988, "data_wait_s": 1.3318000128492713e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 79400, "loss": 0.6087605953216553, "lr": 0.0003, "grad_norm": 0.1344129893234316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11619757036, "tokens_per_second": 149973.354690625, "data_wait_s": 1.0345996997784823e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 79500, "loss": 0.6079615652561188, "lr": 0.0003, "grad_norm": 0.1237358702964263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11634418827, "tokens_per_second": 150010.77367371417, "data_wait_s": 1.6288002370856702e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 79600, "loss": 0.6493849139660597, "lr": 0.0003, "grad_norm": 0.13829361125583936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11649123103, "tokens_per_second": 133717.5246146714, "data_wait_s": 1.4385004760697484e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:52ae7fc5a75a760b57583404d00254f7925887ae40a74e2cf24587ec1dc2cca6
-size 842256
+oid sha256:cb3ec21026275a922dd11846176cce6a98517576e41c4897ca21a9c62f6c5f0b
+size 844374