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": 77194.6200967447,
-      "required_pages_per_second_upper_estimate": 0.005340502934488818
+      "required_token_positions_per_second": 76381.29579934265,
+      "required_pages_per_second_upper_estimate": 0.005284235272422173
     },
     {
       "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": 9649.327512093087,
-      "required_pages_per_second_upper_estimate": 0.0040399719284414945
+      "required_token_positions_per_second": 9547.661974917832,
+      "required_pages_per_second_upper_estimate": 0.003997406691044049
     },
     {
       "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": 19298.655024186173,
-      "required_pages_per_second_upper_estimate": 0.004019400901780276
+      "required_token_positions_per_second": 19095.323949835663,
+      "required_pages_per_second_upper_estimate": 0.0039770524011940875
     },
     {
       "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": 28947.982536279258,
-      "required_pages_per_second_upper_estimate": 0.00173538735203993
+      "required_token_positions_per_second": 28642.9859247535,
+      "required_pages_per_second_upper_estimate": 0.0017171032708818215
     },
     {
       "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": 28947.982536279258,
-      "required_pages_per_second_upper_estimate": 0.002250578313477049
+      "required_token_positions_per_second": 28642.9859247535,
+      "required_pages_per_second_upper_estimate": 0.0022268661684692373
     },
     {
       "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": 28947.982536279258,
-      "required_pages_per_second_upper_estimate": 0.37131363805337614
+      "required_token_positions_per_second": 28642.9859247535,
+      "required_pages_per_second_upper_estimate": 0.3674014690005708
     }
   ],
   "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": 0.05122562764841376,
-    "producer_p95_seconds": 0.15554867399987415,
-    "gpu_update_seconds": 0.9550924650914776,
-    "producer_batches_per_second": 19.52147871888429,
-    "gpu_batches_per_second": 1.0470190442809337,
-    "producer_headroom": 18.644817231850016,
+    "producer_batch_seconds": 0.1329050824296516,
+    "producer_p95_seconds": 0.1619068159998278,
+    "gpu_update_seconds": 0.965262492975859,
+    "producer_batches_per_second": 7.524166734024736,
+    "gpu_batches_per_second": 1.035987627486744,
+    "producer_headroom": 7.2627959392507435,
     "sustainable": true,
-    "buffer_seconds": 60.17082530076309,
+    "buffer_seconds": 60.811537057479114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354959.8007399338,
-    "upload_bytes_per_second": 23802.31602061923,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 503686.30718672473,
+    "upload_bytes_per_second": 3695627.564527411,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.16611019752989,
+    "estimated_full_shards_per_hour": 108.07935630513484,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 23252,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.0272072289999414,
+      "bytes": 7383107,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 689100, "loss": 0.5324979424476624, "lr": 8e-05, "grad_norm": 0.20540835619071118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100865561638, "tokens_per_second": 153379.23205891647, "data_wait_s": 1.6363999748136848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689200, "loss": 0.5125613212585449, "lr": 8e-05, "grad_norm": 0.19679854699535562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100880191325, "tokens_per_second": 152581.70429512163, "data_wait_s": 1.4514000213239342e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689300, "loss": 0.5754213929176331, "lr": 8e-05, "grad_norm": 0.20315760758231538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100894851841, "tokens_per_second": 151678.56312898098, "data_wait_s": 1.902799976960523e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689400, "loss": 0.542589008808136, "lr": 8e-05, "grad_norm": 0.19721698587079897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100909441741, "tokens_per_second": 151855.8219929631, "data_wait_s": 1.6449999748147093e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c277cded59f7e0dc7d9878e724bf52952d1683ee8aacf2b5d0805f25981061b3
-size 91239
+oid sha256:be5a9658c1bb7be90acfb56b27c06dc77387b1bad0aea35d1454d365ecd41ee4
+size 93357