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": 75776.83264349996,
-      "required_pages_per_second_upper_estimate": 0.005242417108753221
+      "required_token_positions_per_second": 75991.40848115578,
+      "required_pages_per_second_upper_estimate": 0.005257261963086789
     },
     {
       "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": 9472.104080437495,
-      "required_pages_per_second_upper_estimate": 0.003965772178453389
+      "required_token_positions_per_second": 9498.926060144473,
+      "required_pages_per_second_upper_estimate": 0.0039770019812500725
     },
     {
       "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": 18944.20816087499,
-      "required_pages_per_second_upper_estimate": 0.003945578967544927
+      "required_token_positions_per_second": 18997.852120288946,
+      "required_pages_per_second_upper_estimate": 0.003956751589604511
     },
     {
       "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": 28416.312241312484,
-      "required_pages_per_second_upper_estimate": 0.001703514529670207
+      "required_token_positions_per_second": 28496.77818043342,
+      "required_pages_per_second_upper_estimate": 0.001708338339856132
     },
     {
       "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": 28416.312241312484,
-      "required_pages_per_second_upper_estimate": 0.002209243286613862
+      "required_token_positions_per_second": 28496.77818043342,
+      "required_pages_per_second_upper_estimate": 0.0022154991594482533
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28416.312241312484,
-      "required_pages_per_second_upper_estimate": 0.3644939423726284
+      "required_token_positions_per_second": 28496.77818043342,
+      "required_pages_per_second_upper_estimate": 0.3655260730420777
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09836509551550421,
-    "producer_p95_seconds": 0.22106178299873136,
-    "gpu_update_seconds": 0.9729622818475548,
-    "producer_batches_per_second": 10.166207787013036,
-    "gpu_batches_per_second": 1.0277890712280269,
-    "producer_headroom": 9.891336726188586,
+    "producer_batch_seconds": 0.08603192614847899,
+    "producer_p95_seconds": 0.2382367329992121,
+    "gpu_update_seconds": 0.9702149423678986,
+    "producer_batches_per_second": 11.623591900919909,
+    "gpu_batches_per_second": 1.0306994422899818,
+    "producer_headroom": 11.277382546258982,
     "sustainable": true,
-    "buffer_seconds": 60.3236614745484,
+    "buffer_seconds": 60.15332642680971,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273884.5668239526,
-    "upload_bytes_per_second": 5096075.578104037,
+    "generation_bytes_per_second": 291111.3394165927,
+    "upload_bytes_per_second": 5100683.334792493,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.76925233401235,
+    "estimated_full_shards_per_hour": 62.46571671365104,
     "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": 0.9322610279996297,
-      "bytes": 6251371,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1527972080002655,
       "bytes": 6634416,
@@ -562,6 +556,12 @@
       "bytes": 6774777,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0792735680006444,
+      "bytes": 7390991,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 460000, "loss": 0.599702775478363, "lr": 0.0003, "grad_norm": 0.16255731056963071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67329946830, "tokens_per_second": 150827.70684953185, "data_wait_s": 1.3354001566767693e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460100, "loss": 0.549582302570343, "lr": 0.0003, "grad_norm": 0.1420083697173705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67344454921, "tokens_per_second": 133765.46454767417, "data_wait_s": 1.1686999641824514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460200, "loss": 0.5797005295753479, "lr": 0.0003, "grad_norm": 0.1720838286779801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67359130439, "tokens_per_second": 152246.32421488495, "data_wait_s": 1.3431999832391739e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 460300, "loss": 0.6818230748176575, "lr": 0.0003, "grad_norm": 0.1774673164541843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67373822275, "tokens_per_second": 152216.01214944094, "data_wait_s": 1.3736000255448744e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac9fe8c853afdbd6fb867d4c6914e8f1ec9a1a48415cadc0a418d8a586ef5ac6
-size 426897
+oid sha256:302bffdd0de53db11915e3f71aa251260974464db98d2717992c2148d681da11
+size 429015