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": 77028.39232071699,
-      "required_pages_per_second_upper_estimate": 0.005329002911241642
+      "required_token_positions_per_second": 76878.95988427367,
+      "required_pages_per_second_upper_estimate": 0.005318664828557471
     },
     {
       "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": 9628.549040089623,
-      "required_pages_per_second_upper_estimate": 0.004031272416117481
+      "required_token_positions_per_second": 9609.86998553421,
+      "required_pages_per_second_upper_estimate": 0.004023451886038144
     },
     {
       "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": 19257.098080179247,
-      "required_pages_per_second_upper_estimate": 0.004010745686273944
+      "required_token_positions_per_second": 19219.73997106842,
+      "required_pages_per_second_upper_estimate": 0.004002964977345748
     },
     {
       "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": 28885.64712026887,
-      "required_pages_per_second_upper_estimate": 0.0017316504390307765
+      "required_token_positions_per_second": 28829.609956602628,
+      "required_pages_per_second_upper_estimate": 0.0017282910966327798
     },
     {
       "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": 28885.64712026887,
-      "required_pages_per_second_upper_estimate": 0.0022457320090667598
+      "required_token_positions_per_second": 28829.609956602628,
+      "required_pages_per_second_upper_estimate": 0.0022413753660731435
     },
     {
       "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": 28885.64712026887,
-      "required_pages_per_second_upper_estimate": 0.37051406626734995
+      "required_token_positions_per_second": 28829.609956602628,
+      "required_pages_per_second_upper_estimate": 0.369795281699858
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.19908104925758607,
-    "producer_p95_seconds": 0.3099698179976258,
-    "gpu_update_seconds": 0.9571535608977089,
-    "producer_batches_per_second": 5.023079814624267,
-    "gpu_batches_per_second": 1.0447644357736137,
-    "producer_headroom": 4.80785873124102,
+    "producer_batch_seconds": 0.13254504507790443,
+    "producer_p95_seconds": 0.31893394800135866,
+    "gpu_update_seconds": 0.9590140151607561,
+    "producer_batches_per_second": 7.544604925912107,
+    "gpu_batches_per_second": 1.0427376286386947,
+    "producer_headroom": 7.235381862800589,
     "sustainable": true,
-    "buffer_seconds": 60.30067433655566,
+    "buffer_seconds": 60.41788295512764,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 562449.5387410478,
-    "upload_bytes_per_second": 5759595.0422652075,
+    "generation_bytes_per_second": 559001.6759720558,
+    "upload_bytes_per_second": 5771072.048088812,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.344288929336436,
+    "estimated_full_shards_per_hour": 119.94874677058463,
     "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": 1.26428677900185,
-      "bytes": 11594708,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.083468554999854,
       "bytes": 6825645,
@@ -562,6 +556,12 @@
       "bytes": 1223987,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.2477019600009953,
+      "bytes": 12351474,
+      "pages": 36,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 587900, "loss": 0.5522806644439697, "lr": 0.0003, "grad_norm": 0.13940387312785324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86050949198, "tokens_per_second": 151149.34521239338, "data_wait_s": 1.1436997738201171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588000, "loss": 0.5519154667854309, "lr": 0.0003, "grad_norm": 0.18406926518733627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86065482954, "tokens_per_second": 150910.2206949209, "data_wait_s": 1.1206000635866076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588100, "loss": 0.5901328325271606, "lr": 0.0003, "grad_norm": 0.16871667434024037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86079988934, "tokens_per_second": 150928.4543544021, "data_wait_s": 1.1067000741604716e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 588200, "loss": 0.5681907534599304, "lr": 0.0003, "grad_norm": 0.17472111764928752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86094666912, "tokens_per_second": 152947.53144042415, "data_wait_s": 1.0006999218603596e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e98dff66483b44eff3ce37304ed5ace7e223c3e40a86af8d7b20d512c4078d53
-size 661443
+oid sha256:4e70f4466ec7ff2816ef8064baa503d9941ad50091587c66d825cd52cead7a25
+size 663561