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": 77408.00321305855,
-      "required_pages_per_second_upper_estimate": 0.005355265273592454
+      "required_token_positions_per_second": 77419.50847651239,
+      "required_pages_per_second_upper_estimate": 0.005356061234414081
     },
     {
       "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": 9676.000401632318,
-      "required_pages_per_second_upper_estimate": 0.0040511393102983
+      "required_token_positions_per_second": 9677.438559564049,
+      "required_pages_per_second_upper_estimate": 0.004051741436992162
     },
     {
       "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": 19352.000803264636,
-      "required_pages_per_second_upper_estimate": 0.004030511420739521
+      "required_token_positions_per_second": 19354.877119128098,
+      "required_pages_per_second_upper_estimate": 0.00403111048148033
     },
     {
       "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": 29028.001204896955,
-      "required_pages_per_second_upper_estimate": 0.0017401843490421286
+      "required_token_positions_per_second": 29032.315678692146,
+      "required_pages_per_second_upper_estimate": 0.0017404429951583303
     },
     {
       "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": 29028.001204896955,
-      "required_pages_per_second_upper_estimate": 0.0022567994129971485
+      "required_token_positions_per_second": 29032.315678692146,
+      "required_pages_per_second_upper_estimate": 0.00225713484435736
     },
     {
       "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": 29028.001204896955,
-      "required_pages_per_second_upper_estimate": 0.37234003161705154
+      "required_token_positions_per_second": 29032.315678692146,
+      "required_pages_per_second_upper_estimate": 0.3723953730543752
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09297908752333228,
-    "producer_p95_seconds": 0.2608476620007423,
-    "gpu_update_seconds": 0.9524596545536813,
-    "producer_batches_per_second": 10.755106622755992,
-    "gpu_batches_per_second": 1.0499132380243401,
-    "producer_headroom": 10.243805138598182,
+    "producer_batch_seconds": 0.07178539703897968,
+    "producer_p95_seconds": 0.24612683599843876,
+    "gpu_update_seconds": 0.9523181101358668,
+    "producer_batches_per_second": 13.930409822167551,
+    "gpu_batches_per_second": 1.0500692881471407,
+    "producer_headroom": 13.266181555264717,
     "sustainable": true,
-    "buffer_seconds": 60.004958236881926,
+    "buffer_seconds": 60.94835904869547,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 515371.43647932884,
-    "upload_bytes_per_second": 3762827.551262156,
+    "generation_bytes_per_second": 516528.784052661,
+    "upload_bytes_per_second": 3736680.551613935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.58671303543949,
+    "estimated_full_shards_per_hour": 110.83505288300393,
     "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": 1.2719860980032536,
-      "bytes": 10158142,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.8902074819998234,
       "bytes": 2207572,
@@ -562,6 +556,12 @@
       "bytes": 2402302,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.039769940001861,
+      "bytes": 6252309,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -239,3 +239,4 @@
 {"step": 548900, "loss": 0.5981613993644714, "lr": 0.0003, "grad_norm": 0.159730107147261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80341651491, "tokens_per_second": 154378.8895892179, "data_wait_s": 1.0288000339642167e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549000, "loss": 0.5882133841514587, "lr": 0.0003, "grad_norm": 0.1836865258532789, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80356352026, "tokens_per_second": 153100.45357766165, "data_wait_s": 1.0129002475878224e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549100, "loss": 0.5335857272148132, "lr": 0.0003, "grad_norm": 0.15731797513821413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80370924075, "tokens_per_second": 151901.74387448313, "data_wait_s": 9.670999133959413e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 549200, "loss": 0.5703690052032471, "lr": 0.0003, "grad_norm": 0.14491462774986033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80385500072, "tokens_per_second": 151527.3272872314, "data_wait_s": 1.442299981135875e-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 sha256:8bb59d40cdf8cfdb96dd497a13864c4ddaa18d3a04e714e5e6ad5ded0de2cf48
-size 513735
+oid sha256:044814b82643aefa7a489e646af873deed93bfe215ad26181b82c47f6b7f959a
+size 515853