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": 75305.0730536308,
-      "required_pages_per_second_upper_estimate": 0.006527252873325981
+      "required_token_positions_per_second": 74975.92749398766,
+      "required_pages_per_second_upper_estimate": 0.006498723370427908
     },
     {
       "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": 9413.13413170385,
-      "required_pages_per_second_upper_estimate": 0.0037550399440337678
+      "required_token_positions_per_second": 9371.990936748458,
+      "required_pages_per_second_upper_estimate": 0.0037386273084204793
     },
     {
       "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": 18826.2682634077,
-      "required_pages_per_second_upper_estimate": 0.004355125831734294
+      "required_token_positions_per_second": 18743.981873496916,
+      "required_pages_per_second_upper_estimate": 0.004336090323619435
     },
     {
       "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": 28239.402395111552,
-      "required_pages_per_second_upper_estimate": 0.0017196314451683882
+      "required_token_positions_per_second": 28115.972810245374,
+      "required_pages_per_second_upper_estimate": 0.0017121152310350216
     },
     {
       "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": 28239.402395111552,
-      "required_pages_per_second_upper_estimate": 0.0021739473680447265
+      "required_token_positions_per_second": 28115.972810245374,
+      "required_pages_per_second_upper_estimate": 0.002164445416926769
     },
     {
       "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": 28239.402395111552,
-      "required_pages_per_second_upper_estimate": 0.35063886654719634
+      "required_token_positions_per_second": 28115.972810245374,
+      "required_pages_per_second_upper_estimate": 0.3491062841954055
     }
   ],
   "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": 0.09255154117215625,
-    "producer_p95_seconds": 0.3633057630067924,
-    "gpu_update_seconds": 0.9790575456648499,
-    "producer_batches_per_second": 10.804790361512055,
-    "gpu_batches_per_second": 1.0213904222768935,
-    "producer_headroom": 10.57851153276522,
+    "producer_batch_seconds": 0.2942828031562499,
+    "producer_p95_seconds": 0.384249770999304,
+    "gpu_update_seconds": 0.9833556244557596,
+    "producer_batches_per_second": 3.3980918669890765,
+    "gpu_batches_per_second": 1.0169260999076017,
+    "producer_headroom": 3.341532749821081,
     "sustainable": true,
-    "buffer_seconds": 60.70156783122069,
+    "buffer_seconds": 60.968048716257094,
     "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": 406008.263519214,
-    "upload_bytes_per_second": 5961746.3880789,
+    "generation_bytes_per_second": 430821.87901259423,
+    "upload_bytes_per_second": 5948481.712472179,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.11992196256938,
+    "estimated_full_shards_per_hour": 92.44434621604319,
     "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": 1.093655648001004,
-      "bytes": 8128170,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1689348799991421,
       "bytes": 10022604,
@@ -562,6 +556,12 @@
       "bytes": 6315821,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1457721929982654,
+      "bytes": 7494949,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -307,3 +307,4 @@
 {"step": 115700, "loss": 0.6948383757844567, "lr": 0.0003, "grad_norm": 0.1684999602518139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16931012208, "tokens_per_second": 148826.94533303508, "data_wait_s": 1.2807002349290997e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115800, "loss": 0.603748444467783, "lr": 0.0003, "grad_norm": 0.1584251483595853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16945791532, "tokens_per_second": 150606.8792175921, "data_wait_s": 1.4125005691312253e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115900, "loss": 0.5962846651673317, "lr": 0.0003, "grad_norm": 0.14476901545958598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16960422649, "tokens_per_second": 148634.60798728914, "data_wait_s": 1.1573996744118631e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 116000, "loss": 0.6011932659894228, "lr": 0.0003, "grad_norm": 0.1298500720331202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16975010873, "tokens_per_second": 148774.85320683385, "data_wait_s": 1.6281002899631858e-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:ff59c688468a6c017b54de0fb57f964365b1bfac448df03b133b082f3a091ace
-size 659325
+oid sha256:d814e208275d60348b4c651f6cd1fede2dec23cc2c9987977bc0364d0f2ee7ce
+size 661443