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": 75170.96359385569,
-      "required_pages_per_second_upper_estimate": 0.006515628605249992
+      "required_token_positions_per_second": 74961.02086788724,
+      "required_pages_per_second_upper_estimate": 0.006497431301858022
     },
     {
       "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": 9396.370449231961,
-      "required_pages_per_second_upper_estimate": 0.003748352660456343
+      "required_token_positions_per_second": 9370.127608485906,
+      "required_pages_per_second_upper_estimate": 0.003737883998917307
     },
     {
       "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": 18792.740898463922,
-      "required_pages_per_second_upper_estimate": 0.004347369865916021
+      "required_token_positions_per_second": 18740.25521697181,
+      "required_pages_per_second_upper_estimate": 0.004335228227219267
     },
     {
       "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": 28189.11134769588,
-      "required_pages_per_second_upper_estimate": 0.0017165689842375076
+      "required_token_positions_per_second": 28110.382825457717,
+      "required_pages_per_second_upper_estimate": 0.001711774830289837
     },
     {
       "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": 28189.11134769588,
-      "required_pages_per_second_upper_estimate": 0.0021700758240001386
+      "required_token_positions_per_second": 28110.382825457717,
+      "required_pages_per_second_upper_estimate": 0.0021640150844237467
     },
     {
       "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": 28189.11134769588,
-      "required_pages_per_second_upper_estimate": 0.35001442005160216
+      "required_token_positions_per_second": 28110.382825457717,
+      "required_pages_per_second_upper_estimate": 0.3490368752928184
     }
   ],
   "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": 0.061092382609359674,
-    "producer_p95_seconds": 0.20927584099990781,
-    "gpu_update_seconds": 0.9808042424246158,
-    "producer_batches_per_second": 16.36865280560845,
-    "gpu_batches_per_second": 1.0195714463142318,
-    "producer_headroom": 16.054444114516357,
+    "producer_batch_seconds": 0.19757269584384574,
+    "producer_p95_seconds": 0.22650405499916815,
+    "gpu_update_seconds": 0.9835511729481333,
+    "producer_batches_per_second": 5.061428127651625,
+    "gpu_batches_per_second": 1.0167239158513353,
+    "producer_headroom": 4.9781735717444295,
     "sustainable": true,
-    "buffer_seconds": 60.809863030326184,
+    "buffer_seconds": 60.98017272278427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355412.2856136534,
-    "upload_bytes_per_second": 5352634.606132625,
+    "generation_bytes_per_second": 404314.99175349623,
+    "upload_bytes_per_second": 5285966.0859254915,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.26320291812135,
+    "estimated_full_shards_per_hour": 86.75658525899567,
     "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": 1.441364390000217,
-      "bytes": 13613764,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.8236200929995903,
       "bytes": 1218836,
@@ -562,6 +556,12 @@
       "bytes": 6097996,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9499315409993869,
+      "bytes": 5779343,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -108,3 +108,4 @@
 {"step": 205800, "loss": 0.5915245972573757, "lr": 0.0003, "grad_norm": 0.14105706562464318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30121589290, "tokens_per_second": 146875.43034318462, "data_wait_s": 1.649799924052786e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205900, "loss": 0.5897836480289698, "lr": 0.0003, "grad_norm": 0.17344364810716098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30136072765, "tokens_per_second": 145956.09273378475, "data_wait_s": 1.3511998986359686e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206000, "loss": 0.5679247565567493, "lr": 0.0003, "grad_norm": 0.1484624834222848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30150532921, "tokens_per_second": 146319.76735478037, "data_wait_s": 1.7637999917496927e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 206100, "loss": 0.6121695507317781, "lr": 0.0003, "grad_norm": 0.14673832847579008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30165066372, "tokens_per_second": 147177.7628433172, "data_wait_s": 2.0351999410195276e-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:a8354c8e15ffa87f004dcde2d65cd38b689917a068b79b542cba7ee64175fe52
-size 234711
+oid sha256:e41f6c045f40fc36748df635adab95a29d7be311bcc69402af3df506196d401d
+size 236829