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": 77606.28866600104,
-      "required_pages_per_second_upper_estimate": 0.005368983121312654
+      "required_token_positions_per_second": 77316.46098038087,
+      "required_pages_per_second_upper_estimate": 0.005348932169528569
     },
     {
       "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": 9700.78608325013,
-      "required_pages_per_second_upper_estimate": 0.004061516557607789
+      "required_token_positions_per_second": 9664.557622547609,
+      "required_pages_per_second_upper_estimate": 0.004046348457647932
     },
     {
       "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": 19401.57216650026,
-      "required_pages_per_second_upper_estimate": 0.004040835828416741
+      "required_token_positions_per_second": 19329.115245095218,
+      "required_pages_per_second_upper_estimate": 0.004025744962505585
     },
     {
       "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": 29102.35824975039,
-      "required_pages_per_second_upper_estimate": 0.0017446419403444585
+      "required_token_positions_per_second": 28993.67286764283,
+      "required_pages_per_second_upper_estimate": 0.0017381264176400796
     },
     {
       "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": 29102.35824975039,
-      "required_pages_per_second_upper_estimate": 0.0022625803461724277
+      "required_token_positions_per_second": 28993.67286764283,
+      "required_pages_per_second_upper_estimate": 0.002254130535770038
     },
     {
       "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": 29102.35824975039,
-      "required_pages_per_second_upper_estimate": 0.37329380395005696
+      "required_token_positions_per_second": 28993.67286764283,
+      "required_pages_per_second_upper_estimate": 0.3718997045656524
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1818507274687562,
-    "producer_p95_seconds": 0.18118723999941722,
-    "gpu_update_seconds": 0.9500261031333135,
-    "producer_batches_per_second": 5.499015670266209,
-    "gpu_batches_per_second": 1.0526026566026616,
-    "producer_headroom": 5.224208428292032,
+    "producer_batch_seconds": 0.10424095839839254,
+    "producer_p95_seconds": 0.19755474099929415,
+    "gpu_update_seconds": 0.9535873611533843,
+    "producer_batches_per_second": 9.59315815361326,
+    "gpu_batches_per_second": 1.0486716170299055,
+    "producer_headroom": 9.147914368831142,
     "sustainable": true,
-    "buffer_seconds": 60.80167060053206,
+    "buffer_seconds": 60.07600375266321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 537992.772335077,
-    "upload_bytes_per_second": 3096600.82148347,
+    "generation_bytes_per_second": 490490.7673123541,
+    "upload_bytes_per_second": 3277721.1260670447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.44072511233553,
+    "estimated_full_shards_per_hour": 105.24790062454193,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 6316246,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5784334970003329,
+      "bytes": 16058487,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 692700, "loss": 0.5638164281845093, "lr": 8e-05, "grad_norm": 0.21678316106732431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101392716786, "tokens_per_second": 153700.706609766, "data_wait_s": 1.4446999557549134e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692800, "loss": 0.5742419362068176, "lr": 8e-05, "grad_norm": 0.21497819229656606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101407438831, "tokens_per_second": 154094.2638992538, "data_wait_s": 1.4145999557513278e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692900, "loss": 0.5348964929580688, "lr": 8e-05, "grad_norm": 0.20250001247449992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101422163664, "tokens_per_second": 153221.58892017504, "data_wait_s": 1.732199962134473e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 693000, "loss": 0.5250784158706665, "lr": 8e-05, "grad_norm": 0.20792641863577016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101436770099, "tokens_per_second": 152125.52605987774, "data_wait_s": 1.6616999346297234e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e24380a1640349ef29ea4b20cba47bb95919d9ffba7ad7b429383fc39067fd4
-size 168270
+oid sha256:e0391dfef879efff05c249fcf0de6f48a0fd91e7ff54fac4de8c241f20c89190
+size 170388