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": 77256.47156568326,
-      "required_pages_per_second_upper_estimate": 0.005344781962625169
+      "required_token_positions_per_second": 77101.28351914833,
+      "required_pages_per_second_upper_estimate": 0.005334045693480004
     },
     {
       "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": 9657.058945710407,
-      "required_pages_per_second_upper_estimate": 0.004043208918246373
+      "required_token_positions_per_second": 9637.660439893541,
+      "required_pages_per_second_upper_estimate": 0.0040350871689477175
     },
     {
       "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": 19314.117891420814,
-      "required_pages_per_second_upper_estimate": 0.00402262140924202
+      "required_token_positions_per_second": 19275.320879787083,
+      "required_pages_per_second_upper_estimate": 0.0040145410148647144
     },
     {
       "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": 28971.176837131217,
-      "required_pages_per_second_upper_estimate": 0.0017367778149603582
+      "required_token_positions_per_second": 28912.981319680624,
+      "required_pages_per_second_upper_estimate": 0.0017332890825486054
     },
     {
       "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": 28971.176837131217,
-      "required_pages_per_second_upper_estimate": 0.0022523815683473413
+      "required_token_positions_per_second": 28912.981319680624,
+      "required_pages_per_second_upper_estimate": 0.002247857122840587
     },
     {
       "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": 28971.176837131217,
-      "required_pages_per_second_upper_estimate": 0.37161114964060515
+      "required_token_positions_per_second": 28912.981319680624,
+      "required_pages_per_second_upper_estimate": 0.37086468002822726
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1216782102503089,
-    "producer_p95_seconds": 0.34166465200542007,
-    "gpu_update_seconds": 0.9543278188328423,
-    "producer_batches_per_second": 8.218398330669574,
-    "gpu_batches_per_second": 1.0478579585189243,
-    "producer_headroom": 7.843046153207367,
+    "producer_batch_seconds": 0.12360201301561347,
+    "producer_p95_seconds": 0.3256195290014148,
+    "gpu_update_seconds": 0.9562486723283852,
+    "producer_batches_per_second": 8.090483120802244,
+    "gpu_batches_per_second": 1.0457530859259485,
+    "producer_headroom": 7.736513742762356,
     "sustainable": true,
-    "buffer_seconds": 60.122652586469066,
+    "buffer_seconds": 60.243666356688266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352085.4503881307,
-    "upload_bytes_per_second": 2956725.255736881,
+    "generation_bytes_per_second": 292849.64254826633,
+    "upload_bytes_per_second": 2950925.4478334687,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.54934152348461,
+    "estimated_full_shards_per_hour": 62.83871610008232,
     "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": 1.2406462889994145,
-      "bytes": 5768030,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 2.4877858859981643,
       "bytes": 7529407,
@@ -562,6 +556,12 @@
       "bytes": 6475828,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.7123713399996632,
+      "bytes": 6324939,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 883200, "loss": 0.5424174666404724, "lr": 3e-05, "grad_norm": 0.3150735655272916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129275704420, "tokens_per_second": 153472.3916388201, "data_wait_s": 1.5719000657554716e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 883300, "loss": 0.5241944193840027, "lr": 3e-05, "grad_norm": 0.3279695984348214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129290365080, "tokens_per_second": 152448.62843261316, "data_wait_s": 1.1127005564048886e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 883400, "loss": 0.5266786813735962, "lr": 3e-05, "grad_norm": 0.35947491459576525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129304983932, "tokens_per_second": 125776.63674631277, "data_wait_s": 8.950999472290277e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 883500, "loss": 0.4863302707672119, "lr": 3e-05, "grad_norm": 0.2961368426973886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129319508864, "tokens_per_second": 149855.1608001854, "data_wait_s": 1.3716999092139304e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74da9cb37fe77631e6af79fb76af0c1b3e0ffe09bfb53fb9bb61f19ac4c34784
-size 872691
+oid sha256:a2a165ecac799751ac25eab5e80148406ebac3e5bed29d6651985ecfb6b6d873
+size 874809