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": 75999.29417555244,
-      "required_pages_per_second_upper_estimate": 0.005257807513722482
+      "required_token_positions_per_second": 75127.57863481477,
+      "required_pages_per_second_upper_estimate": 0.005197500209955528
     },
     {
       "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": 9499.911771944055,
-      "required_pages_per_second_upper_estimate": 0.003977414678196554
+      "required_token_positions_per_second": 9390.947329351846,
+      "required_pages_per_second_upper_estimate": 0.003931793541519507
     },
     {
       "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": 18999.82354388811,
-      "required_pages_per_second_upper_estimate": 0.00395716218515028
+      "required_token_positions_per_second": 18781.89465870369,
+      "required_pages_per_second_upper_estimate": 0.0039117733455375486
     },
     {
       "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": 28499.735315832164,
-      "required_pages_per_second_upper_estimate": 0.0017085156156080024
+      "required_token_positions_per_second": 28172.841988055538,
+      "required_pages_per_second_upper_estimate": 0.0016889188597450043
     },
     {
       "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": 28499.735315832164,
-      "required_pages_per_second_upper_estimate": 0.0022157290637184416
+      "required_token_positions_per_second": 28172.841988055538,
+      "required_pages_per_second_upper_estimate": 0.002190314545335601
     },
     {
       "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": 28499.735315832164,
-      "required_pages_per_second_upper_estimate": 0.3655640039998482
+      "required_token_positions_per_second": 28172.841988055538,
+      "required_pages_per_second_upper_estimate": 0.3613709673818388
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09449509071077955,
-    "producer_p95_seconds": 0.26857112800280447,
-    "gpu_update_seconds": 0.9701142727680349,
-    "producer_batches_per_second": 10.582560347613114,
-    "gpu_batches_per_second": 1.0308063988654572,
-    "producer_headroom": 10.266292835648539,
+    "producer_batch_seconds": 0.2335207904609149,
+    "producer_p95_seconds": 0.28972412099756184,
+    "gpu_update_seconds": 0.9813706409783559,
+    "producer_batches_per_second": 4.282273959531552,
+    "gpu_batches_per_second": 1.0189830001466846,
+    "producer_headroom": 4.202497940510402,
     "sustainable": true,
-    "buffer_seconds": 60.14708491161816,
+    "buffer_seconds": 60.84497974065807,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301004.03364675125,
-    "upload_bytes_per_second": 5137691.342694175,
+    "generation_bytes_per_second": 304933.0984216025,
+    "upload_bytes_per_second": 5096833.164988255,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.58845860530761,
+    "estimated_full_shards_per_hour": 65.43154444204384,
     "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": 1.1848105680001026,
-      "bytes": 6290344,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0463193560008222,
       "bytes": 2621559,
@@ -562,6 +556,12 @@
       "bytes": 10598757,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0763569690025179,
+      "bytes": 2341572,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 739100, "loss": 0.5487536191940308, "lr": 5e-05, "grad_norm": 0.2236926772231008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108184443465, "tokens_per_second": 150100.0227432987, "data_wait_s": 1.2172000424470752e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739200, "loss": 0.5432769060134888, "lr": 5e-05, "grad_norm": 0.26535023741994557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108198931734, "tokens_per_second": 147786.7695840809, "data_wait_s": 1.8216000171378255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739300, "loss": 0.5057598948478699, "lr": 5e-05, "grad_norm": 0.22222499301421045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108213543499, "tokens_per_second": 150266.30178851695, "data_wait_s": 2.1758001821581274e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 739400, "loss": 0.557890772819519, "lr": 5e-05, "grad_norm": 0.22232072734016053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108228161647, "tokens_per_second": 149170.2445662813, "data_wait_s": 1.7991998902289197e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8b98dd2067f9657c37764e98c1234432167f9197e4fb8d0fdeea8c7a8ad0f2d
-size 411288
+oid sha256:8d02e4531245aa3e91e0ee3dc6c33b54cdfb2b2e91ca5567799502bbbf2c5a97
+size 413406