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": 76746.99863903999,
-      "required_pages_per_second_upper_estimate": 0.00530953544341993
+      "required_token_positions_per_second": 76448.66075247723,
+      "required_pages_per_second_upper_estimate": 0.005288895736188249
     },
     {
       "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": 9593.374829879998,
-      "required_pages_per_second_upper_estimate": 0.004016545708823745
+      "required_token_positions_per_second": 9556.082594059653,
+      "required_pages_per_second_upper_estimate": 0.004000932228436205
     },
     {
       "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": 19186.749659759997,
-      "required_pages_per_second_upper_estimate": 0.003996093965513219
+      "required_token_positions_per_second": 19112.165188119307,
+      "required_pages_per_second_upper_estimate": 0.00398055998699525
     },
     {
       "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": 28780.12448964,
-      "required_pages_per_second_upper_estimate": 0.0017253245184483033
+      "required_token_positions_per_second": 28668.24778217896,
+      "required_pages_per_second_upper_estimate": 0.0017186176806618552
     },
     {
       "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": 28780.12448964,
-      "required_pages_per_second_upper_estimate": 0.0022375280886803646
+      "required_token_positions_per_second": 28668.24778217896,
+      "required_pages_per_second_upper_estimate": 0.0022288301667688993
     },
     {
       "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": 28780.12448964,
-      "required_pages_per_second_upper_estimate": 0.3691605352630161
+      "required_token_positions_per_second": 28668.24778217896,
+      "required_pages_per_second_upper_estimate": 0.3677255009835554
     }
   ],
   "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": 0.10498510914847259,
-    "producer_p95_seconds": 0.21438588499950129,
-    "gpu_update_seconds": 0.9606629745452446,
-    "producer_batches_per_second": 9.525160359511316,
-    "gpu_batches_per_second": 1.0409477897005208,
-    "producer_headroom": 9.150468883988593,
+    "producer_batch_seconds": 0.3263542902969334,
+    "producer_p95_seconds": 0.2078722389996983,
+    "gpu_update_seconds": 0.9644119239539579,
+    "producer_batches_per_second": 3.064154600480815,
+    "gpu_batches_per_second": 1.0369013231401534,
+    "producer_headroom": 2.9551072335420745,
     "sustainable": true,
-    "buffer_seconds": 60.52176739635041,
+    "buffer_seconds": 60.757951209099346,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420280.0209768023,
-    "upload_bytes_per_second": 5738374.5109207425,
+    "generation_bytes_per_second": 393197.65477415506,
+    "upload_bytes_per_second": 5782840.946028825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.18230888345767,
+    "estimated_full_shards_per_hour": 84.37106354158868,
     "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": 0.9956328799999028,
-      "bytes": 1276481,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.4530193070004316,
       "bytes": 7690650,
@@ -562,6 +556,12 @@
       "bytes": 11991809,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.543527111000003,
+      "bytes": 7770797,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -147,3 +147,4 @@
 {"step": 539700, "loss": 0.5510649085044861, "lr": 0.0003, "grad_norm": 0.16384735945024742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78993891745, "tokens_per_second": 153414.08955337649, "data_wait_s": 1.0887999451369978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 539800, "loss": 0.6331631541252136, "lr": 0.0003, "grad_norm": 0.18237055770298055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79008423707, "tokens_per_second": 152120.11196001043, "data_wait_s": 8.068000170169398e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 539900, "loss": 0.6281508207321167, "lr": 0.0003, "grad_norm": 0.17473581720289452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79023186415, "tokens_per_second": 154956.58886204657, "data_wait_s": 1.1304000508971512e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 540000, "loss": 0.5902459025382996, "lr": 0.0003, "grad_norm": 0.17200454621042746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79037947186, "tokens_per_second": 154639.35389178383, "data_wait_s": 8.696999429957941e-06, "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 sha256:908f29399a18968c8615dd0351b3ecc5252c4c01e8a7380df365bf9c06009abc
-size 317313
+oid sha256:b892827f2ab9dd037c3f640a949dfd08b9fbfd3fa529c7fb33d3110ef9399739
+size 319431