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": 76849.35053184217,
-      "required_pages_per_second_upper_estimate": 0.005316616384853096
+      "required_token_positions_per_second": 76856.558377742,
+      "required_pages_per_second_upper_estimate": 0.005317115040356953
     },
     {
       "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": 9606.16881648027,
-      "required_pages_per_second_upper_estimate": 0.004021902283844458
+      "required_token_positions_per_second": 9607.06979721775,
+      "required_pages_per_second_upper_estimate": 0.004022279505664623
     },
     {
       "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": 19212.33763296054,
-      "required_pages_per_second_upper_estimate": 0.0040014232655306604
+      "required_token_positions_per_second": 19214.1395944355,
+      "required_pages_per_second_upper_estimate": 0.004001798566584974
     },
     {
       "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": 28818.50644944081,
-      "required_pages_per_second_upper_estimate": 0.0017276254583325026
+      "required_token_positions_per_second": 28821.209391653247,
+      "required_pages_per_second_upper_estimate": 0.0017277874955909844
     },
     {
       "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": 28818.50644944081,
-      "required_pages_per_second_upper_estimate": 0.0022405121172304784
+      "required_token_positions_per_second": 28821.209391653247,
+      "required_pages_per_second_upper_estimate": 0.0022407222590985093
     },
     {
       "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": 28818.50644944081,
-      "required_pages_per_second_upper_estimate": 0.3696528578320033
+      "required_token_positions_per_second": 28821.209391653247,
+      "required_pages_per_second_upper_estimate": 0.3696875282725112
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3372575843359584,
-    "producer_p95_seconds": 0.25396447499952046,
-    "gpu_update_seconds": 0.9593835144963412,
-    "producer_batches_per_second": 2.9650926960440187,
-    "gpu_batches_per_second": 1.0423360260937793,
-    "producer_headroom": 2.844661051538142,
+    "producer_batch_seconds": 0.08967339837491295,
+    "producer_p95_seconds": 0.2607461039988266,
+    "gpu_update_seconds": 0.959293540541258,
+    "producer_batches_per_second": 11.151579154155936,
+    "gpu_batches_per_second": 1.0424337887606063,
+    "producer_headroom": 10.697637849416335,
     "sustainable": true,
-    "buffer_seconds": 60.4411614132695,
+    "buffer_seconds": 60.435493054099254,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335828.13083672104,
-    "upload_bytes_per_second": 3901936.459398772,
+    "generation_bytes_per_second": 355286.66220399394,
+    "upload_bytes_per_second": 3830374.570387652,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.0608991987822,
+    "estimated_full_shards_per_hour": 76.23624705877174,
     "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": 1.5835829299994657,
-      "bytes": 13733184,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.5670746849991701,
       "bytes": 6380319,
@@ -562,6 +556,12 @@
       "bytes": 5996093,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.6912986729985278,
+      "bytes": 6223259,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 816800, "loss": 0.5265446305274963, "lr": 3e-05, "grad_norm": 0.27888402913266097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119556276816, "tokens_per_second": 153480.51494788963, "data_wait_s": 9.121002221945673e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 816900, "loss": 0.5703556537628174, "lr": 3e-05, "grad_norm": 0.28677218077797545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119570995684, "tokens_per_second": 152359.7723636963, "data_wait_s": 1.456300014979206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 817000, "loss": 0.6524937748908997, "lr": 3e-05, "grad_norm": 1.0370803970506586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119585669629, "tokens_per_second": 151843.52674441182, "data_wait_s": 1.5138000890146941e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 817100, "loss": 0.5379834175109863, "lr": 3e-05, "grad_norm": 0.2907010304930393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119600248102, "tokens_per_second": 149959.06256281483, "data_wait_s": 1.432400313206017e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a6353d0b2547c29bf702df1863b162e51d7f9656d28811e7bf048464a0c40c6
-size 362574
+oid sha256:63e4768247f24d77f20f8b2d6df3670a60d144de78625c18da610d75c5e6ea1f
+size 364692