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": 77520.13952026513,
-      "required_pages_per_second_upper_estimate": 0.005363023123517086
+      "required_token_positions_per_second": 77457.63256329442,
+      "required_pages_per_second_upper_estimate": 0.00535869874719773
     },
     {
       "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": 9690.017440033142,
-      "required_pages_per_second_upper_estimate": 0.004057007951567678
+      "required_token_positions_per_second": 9682.204070411803,
+      "required_pages_per_second_upper_estimate": 0.004053736656868935
     },
     {
       "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": 19380.034880066283,
-      "required_pages_per_second_upper_estimate": 0.004036350179628982
+      "required_token_positions_per_second": 19364.408140823605,
+      "required_pages_per_second_upper_estimate": 0.004033095541949559
     },
     {
       "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": 29070.052320099425,
-      "required_pages_per_second_upper_estimate": 0.0017427052491901815
+      "required_token_positions_per_second": 29046.61221123541,
+      "required_pages_per_second_upper_estimate": 0.0017413000504547577
     },
     {
       "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": 29070.052320099425,
-      "required_pages_per_second_upper_estimate": 0.0022600687022408324
+      "required_token_positions_per_second": 29046.61221123541,
+      "required_pages_per_second_upper_estimate": 0.0022582463368787976
     },
     {
       "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": 29070.052320099425,
-      "required_pages_per_second_upper_estimate": 0.37287941817189907
+      "required_token_positions_per_second": 29046.61221123541,
+      "required_pages_per_second_upper_estimate": 0.3725787536234195
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07998961477341027,
-    "producer_p95_seconds": 0.1947022280000965,
-    "gpu_update_seconds": 0.951081879576935,
-    "producer_batches_per_second": 12.501622902332251,
-    "gpu_batches_per_second": 1.051434184031374,
-    "producer_headroom": 11.890067007712215,
+    "producer_batch_seconds": 0.0585760373983959,
+    "producer_p95_seconds": 0.21571295500143606,
+    "gpu_update_seconds": 0.9518493860466655,
+    "producer_batches_per_second": 17.07182739587955,
+    "gpu_batches_per_second": 1.0505863791679473,
+    "producer_headroom": 16.24980842546259,
     "sustainable": true,
-    "buffer_seconds": 60.86924029292384,
+    "buffer_seconds": 60.91836070698659,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455083.79833325115,
-    "upload_bytes_per_second": 5849624.474249078,
+    "generation_bytes_per_second": 456056.23310304433,
+    "upload_bytes_per_second": 5840601.799150988,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.65038931368018,
+    "estimated_full_shards_per_hour": 97.85905117815491,
     "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": 1.0708195489996797,
-      "bytes": 7566500,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1241090929997881,
       "bytes": 6087377,
@@ -562,6 +556,12 @@
       "bytes": 2697855,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9394886710015271,
+      "bytes": 6144198,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 495900, "loss": 0.5551076531410217, "lr": 0.0003, "grad_norm": 0.15962606369740412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72583125799, "tokens_per_second": 153637.7892986949, "data_wait_s": 1.7272001059609465e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 496000, "loss": 0.558561384677887, "lr": 0.0003, "grad_norm": 0.14127708065710512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72598066601, "tokens_per_second": 156266.77688317685, "data_wait_s": 1.3274000593810342e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 496100, "loss": 0.568658709526062, "lr": 0.0003, "grad_norm": 0.15724324929820394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72612733913, "tokens_per_second": 153071.8232457807, "data_wait_s": 9.933000910677947e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 496200, "loss": 0.5768790245056152, "lr": 0.0003, "grad_norm": 0.1476545211548444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72627194811, "tokens_per_second": 150266.67203299157, "data_wait_s": 8.623999747214839e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5dcf1ada51c1b77ed06cf500898f40e71a00cc531c77f94e16668946501cf8eb
-size 236829
+oid sha256:365a27c3d29361bd83d9a24ccae6ae7f12464e99b8a7a58b842beca797d88156
+size 238947