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": 75216.18075821365,
-      "required_pages_per_second_upper_estimate": 0.006519547914454685
+      "required_token_positions_per_second": 75053.57971326381,
+      "required_pages_per_second_upper_estimate": 0.0065054540679870135
     },
     {
       "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": 9402.022594776707,
-      "required_pages_per_second_upper_estimate": 0.003750607385821249
+      "required_token_positions_per_second": 9381.697464157976,
+      "required_pages_per_second_upper_estimate": 0.003742499387329654
     },
     {
       "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": 18804.045189553413,
-      "required_pages_per_second_upper_estimate": 0.0043499849147108464
+      "required_token_positions_per_second": 18763.394928315953,
+      "required_pages_per_second_upper_estimate": 0.004340581192193723
     },
     {
       "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": 28206.06778433012,
-      "required_pages_per_second_upper_estimate": 0.0017176015422649853
+      "required_token_positions_per_second": 28145.09239247393,
+      "required_pages_per_second_upper_estimate": 0.0017138884608141
     },
     {
       "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": 28206.06778433012,
-      "required_pages_per_second_upper_estimate": 0.002171381177430663
+      "required_token_positions_per_second": 28145.09239247393,
+      "required_pages_per_second_upper_estimate": 0.002166687122974886
     },
     {
       "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": 28206.06778433012,
-      "required_pages_per_second_upper_estimate": 0.35022496224505656
+      "required_token_positions_per_second": 28145.09239247393,
+      "required_pages_per_second_upper_estimate": 0.3494678519497117
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10476856211718655,
-    "producer_p95_seconds": 0.18885634199978085,
-    "gpu_update_seconds": 0.980214619471341,
-    "producer_batches_per_second": 9.544847994396184,
-    "gpu_batches_per_second": 1.0201847433568476,
-    "producer_headroom": 9.355999544738848,
+    "producer_batch_seconds": 0.29354809504688006,
+    "producer_p95_seconds": 0.19582203100071638,
+    "gpu_update_seconds": 0.9823382213303072,
+    "producer_batches_per_second": 3.4065967958003562,
+    "gpu_batches_per_second": 1.0179793255379748,
+    "producer_headroom": 3.3464302371760457,
     "sustainable": true,
-    "buffer_seconds": 60.773306407223146,
+    "buffer_seconds": 60.90496972247905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364056.88282280555,
-    "upload_bytes_per_second": 5423901.113376007,
+    "generation_bytes_per_second": 364208.3774637217,
+    "upload_bytes_per_second": 5263876.007180285,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.11813224328161,
+    "estimated_full_shards_per_hour": 78.15063946660746,
     "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": [
@@ -274,6 +274,18 @@
       "bytes": 2653813,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.7666170649999913,
+      "bytes": 10799529,
+      "pages": 36,
+      "shards": 3
+    },
+    {
+      "seconds": 0.9170148619996326,
+      "bytes": 257577,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 199800, "loss": 0.5837657302618027, "lr": 0.0003, "grad_norm": 0.13412700037772263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29242595259, "tokens_per_second": 146476.28526405295, "data_wait_s": 1.7119000403909013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199900, "loss": 0.5994546860456467, "lr": 0.0003, "grad_norm": 0.1577259501370254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29257211636, "tokens_per_second": 147569.86863726124, "data_wait_s": 1.5909000467217993e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 200000, "loss": 0.5858247317373753, "lr": 0.0003, "grad_norm": 0.1438895802713997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29271869223, "tokens_per_second": 147301.7738234091, "data_wait_s": 1.8273999558005016e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 200100, "loss": 0.7128876578062773, "lr": 0.0003, "grad_norm": 0.224431648026927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29286561363, "tokens_per_second": 136632.69797399724, "data_wait_s": 2.6620999960869085e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3daceebde873f2d889ab7df0a2ac2fc37cda3964935c896aba7a797b0e0ec90d
-size 106848
+oid sha256:6a88a9df1cd961b1ee96c64f2f45e3d4f092b924f3194156ed4b48f0ef089c31
+size 108966