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": 77099.05540407023,
-      "required_pages_per_second_upper_estimate": 0.0053338915473089615
+      "required_token_positions_per_second": 76946.22215118262,
+      "required_pages_per_second_upper_estimate": 0.005323318188252193
     },
     {
       "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": 9637.381925508778,
-      "required_pages_per_second_upper_estimate": 0.004034970560790859
+      "required_token_positions_per_second": 9618.277768897828,
+      "required_pages_per_second_upper_estimate": 0.004026972049358009
     },
     {
       "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": 19274.763851017557,
-      "required_pages_per_second_upper_estimate": 0.004014425000461858
+      "required_token_positions_per_second": 19236.555537795655,
+      "required_pages_per_second_upper_estimate": 0.004006467216438716
     },
     {
       "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": 28912.145776526333,
-      "required_pages_per_second_upper_estimate": 0.001733238993012306
+      "required_token_positions_per_second": 28854.833306693483,
+      "required_pages_per_second_upper_estimate": 0.0017298031979569011
     },
     {
       "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": 28912.145776526333,
-      "required_pages_per_second_upper_estimate": 0.0022477921630355063
+      "required_token_positions_per_second": 28854.833306693483,
+      "required_pages_per_second_upper_estimate": 0.0022433363705969154
     },
     {
       "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": 28912.145776526333,
-      "required_pages_per_second_upper_estimate": 0.37085396257778036
+      "required_token_positions_per_second": 28854.833306693483,
+      "required_pages_per_second_upper_estimate": 0.3701188197521002
     }
   ],
   "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": 0.10851886896870155,
-    "producer_p95_seconds": 0.32683736699982546,
-    "gpu_update_seconds": 0.9562763073243533,
-    "producer_batches_per_second": 9.214987305925707,
-    "gpu_batches_per_second": 1.045722865181074,
-    "producer_headroom": 8.812074032951427,
+    "producer_batch_seconds": 0.11970875134363723,
+    "producer_p95_seconds": 0.32819900500180665,
+    "gpu_update_seconds": 0.95817569646422,
+    "producer_batches_per_second": 8.353608142894993,
+    "gpu_batches_per_second": 1.0436499315210317,
+    "producer_headroom": 8.00422430030759,
     "sustainable": true,
-    "buffer_seconds": 60.24540736143426,
+    "buffer_seconds": 60.36506887724586,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344451.52717153495,
-    "upload_bytes_per_second": 5116333.810621351,
+    "generation_bytes_per_second": 345071.01185213745,
+    "upload_bytes_per_second": 5117208.817661999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.91127930984055,
+    "estimated_full_shards_per_hour": 74.04420630143254,
     "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": 1.485799860998668,
-      "bytes": 7350955,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3855582350006443,
       "bytes": 11926342,
@@ -562,6 +556,12 @@
       "bytes": 10987528,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2853161590028321,
+      "bytes": 6397328,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 475800, "loss": 0.6057299375534058, "lr": 0.0003, "grad_norm": 0.1476927660982085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69641974503, "tokens_per_second": 153646.5589215426, "data_wait_s": 1.682799484115094e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475900, "loss": 0.6367251873016357, "lr": 0.0003, "grad_norm": 0.1925758467650204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69656760288, "tokens_per_second": 153248.00235855463, "data_wait_s": 1.4907003787811846e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 476000, "loss": 0.6127805709838867, "lr": 0.0003, "grad_norm": 0.16427813440189706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69671513430, "tokens_per_second": 153968.5091303614, "data_wait_s": 8.830000297166407e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 476100, "loss": 0.640205979347229, "lr": 0.0003, "grad_norm": 0.18893404899756797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69686162963, "tokens_per_second": 152610.91474064012, "data_wait_s": 1.0691001079976559e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be794f6c0719e0ab210bf8ed5947ef6a33acc3388f9b6f7a96538af1c26615bf
-size 763890
+oid sha256:4c129d08d6e86831ed2a9904b0ed2f34ec288d36da74443b24679e5220d4db8a
+size 766008