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": 75436.29306341436,
-      "required_pages_per_second_upper_estimate": 0.006538626691199945
+      "required_token_positions_per_second": 75037.06245120552,
+      "required_pages_per_second_upper_estimate": 0.0065040223935744195
     },
     {
       "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": 9429.536632926794,
-      "required_pages_per_second_upper_estimate": 0.003761583147010848
+      "required_token_positions_per_second": 9379.63280640069,
+      "required_pages_per_second_upper_estimate": 0.003741675764480888
     },
     {
       "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": 18859.07326585359,
-      "required_pages_per_second_upper_estimate": 0.004362714691701827
+      "required_token_positions_per_second": 18759.26561280138,
+      "required_pages_per_second_upper_estimate": 0.004339625947722898
     },
     {
       "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": 28288.609898780385,
-      "required_pages_per_second_upper_estimate": 0.0017226279239700008
+      "required_token_positions_per_second": 28138.89841920207,
+      "required_pages_per_second_upper_estimate": 0.00171351128033911
     },
     {
       "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": 28288.609898780385,
-      "required_pages_per_second_upper_estimate": 0.0021777354978922414
+      "required_token_positions_per_second": 28138.89841920207,
+      "required_pages_per_second_upper_estimate": 0.0021662102937664027
     },
     {
       "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": 28288.609898780385,
-      "required_pages_per_second_upper_estimate": 0.3512498590558425
+      "required_token_positions_per_second": 28138.89841920207,
+      "required_pages_per_second_upper_estimate": 0.3493909435315702
     }
   ],
   "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": 0.07990082986722769,
-    "producer_p95_seconds": 0.33423136100100237,
-    "gpu_update_seconds": 0.9773544935197398,
-    "producer_batches_per_second": 12.5155145655147,
-    "gpu_batches_per_second": 1.0231702075658415,
-    "producer_headroom": 12.232094399317544,
+    "producer_batch_seconds": 0.12852979753910176,
+    "producer_p95_seconds": 0.34119676499904017,
+    "gpu_update_seconds": 0.9825544549794075,
+    "producer_batches_per_second": 7.780297013972785,
+    "gpu_batches_per_second": 1.0177552958334082,
+    "producer_headroom": 7.6445654921419415,
     "sustainable": true,
-    "buffer_seconds": 60.595978598223866,
+    "buffer_seconds": 60.91837620872327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423678.7651222719,
-    "upload_bytes_per_second": 4186789.304258927,
+    "generation_bytes_per_second": 491616.0744627215,
+    "upload_bytes_per_second": 4142906.2818062943,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.91160025836103,
+    "estimated_full_shards_per_hour": 105.489365343201,
     "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": 1.002836310002749,
-      "bytes": 6201464,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0725969490013085,
       "bytes": 8132406,
@@ -562,6 +556,12 @@
       "bytes": 10022604,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8061036489998514,
+      "bytes": 878150,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 108200, "loss": 0.6171077489852905, "lr": 0.0003, "grad_norm": 0.141955425435858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15833177836, "tokens_per_second": 147043.92650415594, "data_wait_s": 1.4761000784346834e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108300, "loss": 0.6314911562949419, "lr": 0.0003, "grad_norm": 0.13984104643722026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15847858078, "tokens_per_second": 149083.8728009691, "data_wait_s": 1.6496000171173364e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108400, "loss": 0.5929444441571832, "lr": 0.0003, "grad_norm": 0.15431779799373826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15862485009, "tokens_per_second": 148121.56615737014, "data_wait_s": 1.4886998542351648e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 108500, "loss": 0.5959181506186724, "lr": 0.0003, "grad_norm": 0.1446121997830395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15876967855, "tokens_per_second": 146694.06227859564, "data_wait_s": 1.4485998690361157e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02d0da809916e1b0126785e8cc1298d0e5321de76f27ed91065ea9a805efa20b
-size 498909
+oid sha256:277e76dbe5eef42cccae0f0662e1c694678aad703b131c4254acefa34e68a4ef
+size 501027