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": 77052.00807466944,
-      "required_pages_per_second_upper_estimate": 0.005330636703896171
+      "required_token_positions_per_second": 76789.86234979887,
+      "required_pages_per_second_upper_estimate": 0.0053125008543877415
     },
     {
       "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": 9631.50100933368,
-      "required_pages_per_second_upper_estimate": 0.00403250834399584
+      "required_token_positions_per_second": 9598.73279372486,
+      "required_pages_per_second_upper_estimate": 0.004018788976398581
     },
     {
       "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": 19263.00201866736,
-      "required_pages_per_second_upper_estimate": 0.004011975320963691
+      "required_token_positions_per_second": 19197.46558744972,
+      "required_pages_per_second_upper_estimate": 0.003998325810652971
     },
     {
       "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": 28894.503028001043,
-      "required_pages_per_second_upper_estimate": 0.001732181337177132
+      "required_token_positions_per_second": 28796.198381174574,
+      "required_pages_per_second_upper_estimate": 0.0017262881237023898
     },
     {
       "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": 28894.503028001043,
-      "required_pages_per_second_upper_estimate": 0.0022464205169399148
+      "required_token_positions_per_second": 28796.198381174574,
+      "required_pages_per_second_upper_estimate": 0.002238777763045601
     },
     {
       "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": 28894.503028001043,
-      "required_pages_per_second_upper_estimate": 0.37062766034300537
+      "required_token_positions_per_second": 28796.198381174574,
+      "required_pages_per_second_upper_estimate": 0.36936671388482156
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.108132203375078,
-    "producer_p95_seconds": 0.23523941399980686,
-    "gpu_update_seconds": 0.9568602018593958,
-    "producer_batches_per_second": 9.247938808120848,
-    "gpu_batches_per_second": 1.0450847449363803,
-    "producer_headroom": 8.848984594721854,
+    "producer_batch_seconds": 0.16438445637510313,
+    "producer_p95_seconds": 0.22943706799924257,
+    "gpu_update_seconds": 0.9601267373569281,
+    "producer_batches_per_second": 6.0833002222432455,
+    "gpu_batches_per_second": 1.0415291659857702,
+    "producer_headroom": 5.840739194745083,
     "sustainable": true,
-    "buffer_seconds": 60.28219271714193,
+    "buffer_seconds": 60.48798445348647,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 562316.0675373055,
-    "upload_bytes_per_second": 5159603.732273359,
+    "generation_bytes_per_second": 578662.3593807718,
+    "upload_bytes_per_second": 3751677.9186152853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.32996902270019,
+    "estimated_full_shards_per_hour": 62.083735876404596,
     "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": 1.3427877739995893,
-      "bytes": 6044460,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2501696180006547,
       "bytes": 2425262,
@@ -562,6 +556,12 @@
       "bytes": 838514,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 34.2643265609986,
+      "bytes": 13380891,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -235,3 +235,4 @@
 {"step": 548500, "loss": 0.5708064436912537, "lr": 0.0003, "grad_norm": 0.1822723920452582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80282930148, "tokens_per_second": 151586.76880784568, "data_wait_s": 1.222100036102347e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 548600, "loss": 0.5830158591270447, "lr": 0.0003, "grad_norm": 0.17392310441937436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80297517543, "tokens_per_second": 151776.5463391494, "data_wait_s": 2.0172999938949943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 548700, "loss": 0.6424188613891602, "lr": 0.0003, "grad_norm": 0.15790707078237418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80312124169, "tokens_per_second": 152152.6743513665, "data_wait_s": 1.3599001249531284e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 548800, "loss": 0.596732497215271, "lr": 0.0003, "grad_norm": 0.16276950425080294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80326866000, "tokens_per_second": 153501.53632570375, "data_wait_s": 2.183499964303337e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:839e0d1a7cf4f9f1922a8e361f2ac280e9cba30d0b38fb83de6bd1cc41a47a30
-size 505263
+oid sha256:d4d07bb1f7b001e9d4287012d4a79ffa71cc0f535a0414d85c0af4f8f19516b0
+size 507381